Download Estimating the Economically Optimal Reserve

Estimating the Economically Optimal
Reserve Margin in ERCOT
January 31, 2014
PREPARED FOR
The Public Utility Commission of Texas
PREPARED BY
Samuel A. Newell
Kathleen Spees
Johannes P. Pfeifenberger
Ioanna Karkatsouli
The Brattle Group
Nick Wintermantel
Kevin Carden
Astrape Consulting
This report was prepared for the Electric Reliability Council of Texas (ERCOT). All results and
any errors are the responsibility of the authors and do not represent the opinion of The Brattle
Group, Inc. or its clients. Opinions expressed in this report, as well as any errors or omissions,
are the authors’ alone. The examples, facts, and requirements summarized in this report
represent our interpretations. Nothing herein is intended to provide a legal opinion.
The authors would like to thank the ERCOT staff for their input, cooperation, and support. In
particular, we would like to acknowledge the analytical, technical, and conceptual contributions
of John Dumas, Warren Lasher, Resmi Surendran, Julia Matevosyan, Kevin Hanson, Pete
Warnken, Calvin Opheim, and Mark Patterson. We are also grateful to Dan Jones and Beth
Garza of Potomac Economics for their helpful comments.
Copyright © 2014 The Brattle Group, Inc.
i | brattle.com
Table of Contents
Executive Summary ...................................................................................................................... iv I. Motivation and Context ......................................................................................................... 1 II. Study Assumptions and Approach .......................................................................................... 3 A. SERVM Probabilistic Modeling Framework ...................................................................... 4 B. Load Modeling ..................................................................................................................... 5 1. Peak Load and Regional Diversity.............................................................................. 6 2. Load Shapes and Weather Years................................................................................. 7 3. Load Forecast Uncertainty and Forward Period........................................................ 8 C. Generation Resources .......................................................................................................... 9 1. Conventional Generation Outages ........................................................................... 10 2. Private Use Networks ................................................................................................ 11 3. Intermittent Wind and Solar .................................................................................... 12 4. Hydroelectric ............................................................................................................. 13 5. Marginal Resource Technology ................................................................................ 14 6. Fuel Prices .................................................................................................................. 16 D. Demand-Side Resources .................................................................................................... 17 1. Costs and Modeling of Demand Resources .............................................................. 18 2. Emergency Response Service.................................................................................... 20 3. Non-Controllable Load Resources ............................................................................ 21 4. Price-Responsive Demand ........................................................................................ 22 E. ERCOT and External Systems’ Resource Overview ........................................................ 23 1. System Topology and Intertie Availability .............................................................. 24 2. Resource Mix Across Reserve Margins .................................................................... 25 3. Supply Curves ............................................................................................................ 27 F. Scarcity Conditions ............................................................................................................ 28 1. Market Parameters .................................................................................................... 28 2. Emergency Procedures and Marginal Costs............................................................. 29 3. Emergency Generation ............................................................................................. 31 4. Power Balance Penalty Curve .................................................................................. 32 5. Operating Reserves Demand Curve ......................................................................... 34 G. Reserve Margin Accounting .............................................................................................. 37 III. Reliability and System Costs Results .................................................................................... 40 A. System Reliability .............................................................................................................. 40 1. Physical Reliability Metrics ...................................................................................... 40 2. Emergency Event Frequency .................................................................................... 45 B. Economically Optimal Reserve Margin ............................................................................ 46 1. System Cost-Minimizing Reserve Margin ............................................................... 46 2. Exposure to Extreme Shortage Events ..................................................................... 49 ii | brattle.com
C. Sensitivity to System Conditions and Study Assumptions .............................................. 51 1. Marginal Resource Technology Type and Cost of New Entry ............................... 51 2. Forward Period and Load Forecast Uncertainty...................................................... 53 3. Probability Weighting of Weather Years ................................................................ 56 4. Energy Prices Always Equal to Marginal Cost ........................................................ 57 D. Sensitivity of Economic Reserve Margin to Study Assumptions .................................... 59 IV. Comparison of Energy-Only and Capacity Market Designs ................................................. 60 A. Energy-Only Market Results............................................................................................. 61 1. Equilibrium Reserve Margin..................................................................................... 61 2. Volatility in Realized Prices and Energy Margins ................................................... 63 3. Year-to-Year Reserve Margin Variability ................................................................ 64 B. Capacity Market Results .................................................................................................... 66 1. Capacity Prices at Higher Reserve Margins ............................................................. 66 2. Capacity Price Volatility ........................................................................................... 68 C. Implications for Customers and Suppliers ........................................................................ 69 1. Total Customer Costs and Volatility ........................................................................ 70 2. Supplier Net Revenues and Volatility ...................................................................... 73 V. Policy Implications............................................................................................................... 75 List of Acronyms ......................................................................................................................... 81 Bibliography ................................................................................................................................ 84 iii | brattle.com
Executive Summary
We have been asked by the Public Utility Commission of Texas (PUCT), in close coordination
with the Electric Reliability Council of Texas (ERCOT), to estimate the economically optimal
reserve margin for ERCOT’s wholesale electric market. This study contributes to an already
substantial body of technical work, regulatory proceedings, and market design revisions related
to the policy framework for resource adequacy in Texas.
We provide the following new information to help the Commission and stakeholders evaluate
their options for addressing the resource adequacy challenges facing Texas:
(a) An estimate of the “economically optimal” reserve margin that would minimize system
costs in ERCOT;
(b) A comparison between that optimum and the “equilibrium” reserve margin that ERCOT’s
current energy-only market design will likely support;
(c) Estimates of the system cost and customer cost implications of mandating a higher reserve
margin, such as one based on the traditional 1-in-10 loss of load event (LOLE) standard or
alternative physical reliability standards;
(d) A comparison of prices, customer costs, and supplier net revenues that energy-only and
capacity market design options would produce—both on average across all years and for
the highest-cost years with significant scarcity events; and
(e) The sensitivity of our reliability and economic results to a number of study assumptions.
Our analyses of viable market designs represent expected long-term average conditions at the
equilibrium point where suppliers are earning adequate returns on average to support continued
investment in new generation. Our results do not describe today’s market conditions or the
reserve margins and prices that may be realized in any of the next few years.
To undertake this analysis, we implement a series of economic and reliability modeling
simulations of the ERCOT system using the Strategic Energy Risk Valuation Model (SERVM).
Like other reliability modeling tools, SERVM probabilistically evaluates resource adequacy
conditions by simulating ERCOT’s generation outages, weather and other load uncertainty,
intertie availability, demand-side resources, and other factors. Unlike other reliability modeling
tools, SERVM also simulates the economic implications associated with all possible outcomes,
including hourly generation dispatch, import-export dynamics, ancillary services, demand
response, and individual emergency procedures. SERVM estimates hourly and annual
production costs, customer costs, market prices, net import costs, load shed costs, and generator
energy margins as a function of the planning reserve margin under a wide range of uncertainties.
We simulated these uncertainties probabilistically for a Base Case, a number of sensitivity cases,
and a range of planning reserve margin levels. The results for each case and simulated reserve
iv | brattle.com
margin level reflect the probability-weighted outcomes for 7,500 full annual simulations (for all
hours of the year).
The “economically optimal” reserve margin minimizes total system costs by weighing:
(1) increasing capital costs of building more generation plants to achieve the higher reserve
margins, against (2) decreasing scarcity-event-related costs as higher reserve margins help to
avoid load shedding, reserve shortages, demand-response calls, and other emergency event costs.
Figure ES-1 summarizes these individual cost components across a range of planning reserve
margins. The minimum system cost (reflecting the risk-neutral, probability-weighted-average
cost of 7,500 simulations) occurs at a reserve margin of 10.2% in our Base Case. This risk-neutral,
economically optimal reserve margin is substantially below the 14.1% reserve margin needed to
meet the traditional 1-in-10 (0.1 LOLE) target in our Base Case simulations, meaning that the
traditional 1-in-10 target is higher than economically optimal, at least before accounting for risk
aversion and other considerations.
Figure ES‐1 Total System Costs across Planning Reserve Margins Economically Optimal Reserve Margin at 10.2%
$36,000
Total System Costs ($Million/Year)
Firm Load Shedding
Regulation Shortages
Non‐Spinning Reserve Shortages
Spinning Reserve Shortages
Price‐Responsive Demand
TDSP Load Management
Non‐Controllable LRs
30‐Minute ERS
10‐Minute ERS
Emergency Generation
External System Costs (Above Baseline)
Production Costs (Above $10B/Yr Baseline)
Marginal CC Capital Costs
$35,800
$35,600
$35,400
$35,200
$35,000
17.5%
16.5%
15.6%
14.6%
13.7%
12.7%
11.8%
10.8%
9.8%
8.9%
7.9%
7.0%
6.0%
$34,800
ERCOT Reserve Margin (% ICAP)
Notes: Total system costs include a large baseline of total system costs that do not change across reserve margins, including $15.2B/year in transmission and distribution, $9.6B/year in fixed costs for generators other than the marginal unit, and $10B/year in production costs. However, we also find that the total system cost curve is relatively flat near the minimum, with
only modest average annual cost variation between reserve margins of 8% and 14%. For
example, increasing the reserve margin from the 10.2% optimum to the 14.1% needed to meet
the 1-in-10 standard, would increase total system costs by approximately $100 million per year
v | brattle.com
on a long-term average basis. This compares to total ERCOT system-wide costs of more than
$35 billion per year including all reliability-related costs, production costs, fleet-wide fixed costs,
and transmission and distribution (T&D) costs.
Our estimates of the economically optimal reserve margin and reliability-based reserve margin
targets are summarized in Table ES-1. Although the 0.1 LOLE target is the traditional metric
used in Texas and most of North America, we also report reserve margins that would be needed
to meet alternative reliability standards: (a) a 9.1% reserve margin would meet the 2.4 loss of
load hours (LOLH) standard that the Southwest Power Pool (SPP) uses,1 and (b) a 9.6% reserve
margin would meet the a 0.001% normalized expected unserved energy (EUE) standard used in
some international markets. To make the implications of the 0.001% normalized EUE standard
more tangible, this means one hundred thousandth of the energy demanded each year would be
shed on average due to resource inadequacy, corresponding to 0.7 events per year, each shedding
about 1,700 MW for 2.8 hours on average in our ERCOT simulations. If the Commission does
implement a reliability-based standard, we would recommend defining the standard in terms of
normalized EUE, although not necessarily at the 0.001% level. Unlike LOLE, normalized EUE is
a more robust metric in terms of its comparability across system sizes, outage durations, and
outage event magnitude, as explained in a recent North American Electric Reliability
Corporation (NERC) task force whitepaper.
Table ES‐1 Economically Optimal and Reliability‐Based Reserve Margin Targets Reserve Margin Target
Base Case
(%)
Sensitivity Cases
(%)
Economically‐Optimal 10.2%
9.3%‐11.5%
Reliability‐Based
0.1 LOLE
2.4 LOLH
0.001% EUE
14.1%
9.1%
9.6%
12.6%‐16.1%
8.2%‐11.1%
8.4%‐11.6%
Notes: This reported sensitivity range reflects a subset of the sensitivities we examined in this study, reflecting the following cases: Equal 1/15 Chance of 2011 Weather, No Non‐Weather Load Forecast Error, Perfect Energy Price, and Combustion Turbine as the Marginal Technology. Table ES-1 also shows that both economically-based and reliability-based reserve margin
estimates are sensitive to study assumptions, particularly those that drive the likelihood and
severity of scarcity events. Two of the most important of these assumptions are the likelihood of
extreme 2011 weather recurring and the magnitude of non-weather load forecast error (LFE).
The economically optimal reserve margin also depends on our economic assumptions regarding
the estimated capital cost of building new generation, the value of lost load (VOLL), the cost of
1
SPP’s LOLH methodology yields a minimum reserve margin requirement of 12% capacity margin
(13.6% reserve margin). Actual SPP reserve margins are currently much higher, however.
vi | brattle.com
dispatching demand response, natural gas prices, and other factors. We show a range of results
based on some of these factors in Table ES-2, and provide a more comprehensive sensitivity
analysis in the body of this report.
This study also compares the long-term implications of maintaining ERCOT’s current energyonly market design to those of implementing a capacity market. Under the energy-only market
construct, the average reserve margin is determined solely by market forces. If reserve margins
are low and prices are high, suppliers will build because they expect to earn more than their
investment costs. Those newly built plants will increase the average reserve margin and thereby
suppress energy prices. Suppliers will continue to build until market prices drop to an
“equilibrium” level where they expect to earn an adequate return on capital, no more and no less.
We estimate that the current energy-only market design will yield an equilibrium reserve
margin of 11.5% under our Base Case assumptions, with a range of 9% to 13% in our sensitivity
cases.
The 11.5% equilibrium reserve margin that the current energy-only market design is likely to
achieve slightly exceeds the 10.2% risk-neutral economically optimal reserve margin. This
important finding suggests that the current market design will support sufficient reserve margins
from an economic perspective, unless political perceptions or reactions have adverse economic
consequences for which we have not accounted. Some considerations of risk aversion are
discussed below.
The reason the energy-only market equilibrium slightly exceeds the cost-minimizing
economically optimal reserve margin is that the current market design occasionally yields energy
prices above marginal system costs, creating additional incentives to invest that raise reserve
margins somewhat above the cost-minimizing level. Specifically, this discrepancy is introduced
by setting administratively-determined scarcity prices as if load would be shed (or other
emergency actions taken at an equivalent cost) at an operating reserve level of X = 2,000 MW.
This is above our X = 1,150 MW estimated level at which load is shed, with prior emergency
actions incurring costs below the value of lost load. Under an alternative “Perfect Energy Price”
case, we illustrate that the energy-only market equilibrium and economically optimal reserve
margins would both be equal to 9.3% if prices were always set equal to our estimates of marginal
system costs. This sensitivity case has a lower optimum reserve margin and lower total system
costs because the “perfect” energy prices also result in more efficient system dispatch, with
respect to high-cost price-responsive demand.
As an alternative to the current energy-only market design, the Commission could opt to
mandate a higher reserve margin. This would make capacity a valuable new product because, at
higher reserve margins, the market would produce lower energy market prices and lower
supplier energy margins. In that case, “capacity payments” would make up the difference needed
to allow suppliers to invest up to the mandated higher reserve margin. On a long-term average
basis, the resulting capacity prices will equilibrate at the Net Cost of New Entry (Net CONE),
which is gross CONE net of energy-market margins. This equilibrium capacity price increases
with the mandated reserve margins as suppliers earn less in the energy market, such that total
vii | brattle.com
net revenues from capacity and energy are equal to gross CONE in equilibrium, consistent with
the long-run marginal cost of supply.
At the reserve margin needed to maintain 0.1 LOLE, which is 14.1% in our Base Case, we
estimate long-run equilibrium capacity prices of $39/kW-year in the Base Case and between
$30–$60/kW-year in our sensitivity cases (25%–49% of total supplier net revenues). These and
other metrics describing the differences between the energy-only and capacity-market designs
are shown in Table ES-2.
One of the most important insights is that the net increase in customer cost under a capacity
market design would be much lower than implied by our estimates of capacity prices. Although
multiplying our estimated capacity price by projected peak load (plus reserve margin) suggests
roughly $3.2 billion per year in capacity payments, these costs would be largely offset by a $2.8
billion reduction in annual energy-market costs, for a net customer cost increase of about $400
million relative to the energy-only market design at the 11.5% equilibrium reserve margin. This
$400 million per year represents only about a 1% increase in customer rates, from about
10.1 ¢/kWh in the 11.5% energy-only equilibrium to 10.2 ¢/kWh with a 14.1% reserve margin
mandate.
However, these estimates describe only long-term average prices at equilibrium. They do not
describe this year or the next few years. The actual near-term price impacts of implementing a
capacity market would be affected by at least two important dynamics. First, capacity prices
could be temporarily lower than estimated if some low-cost capacity is available. Other regions
have experienced capacity prices below Net CONE for many years due to the entry of demand
response, generation uprates, and other low-cost sources of capacity. Second, even without such
resources, prices would not be expected to reach equilibrium pricing until the reserve margin
falls to the required reserve margin. But herein lies an important cost difference from
maintaining an energy-only market. Mandating a reserve margin could cause the market to
reach equilibrium as soon as reserve margins fall from their current levels to 14.1% (or whatever
level is mandated), whereas the current energy-only market design may take several years to
reach its 11.5% long-run equilibrium level. Under either market design, long-term equilibrium
prices would be several billion dollars higher than currently-depressed wholesale prices, but a
capacity requirement would cause prices to reach equilibrium prices sooner, and at a slightly
higher ultimate level (e.g., $400 million higher on average to support a 0.1 LOLE).
viii | brattle.com
Table ES‐2 Comparison of Energy‐Only and Capacity Market Outcomes Energy‐Only Market
Base Case Sensitivity Cases
Equilibrium Reserve Margin
(%)
11.5%
Realized Reliability
Loss of Load Events
Loss of Load Hours
Normalized EUE
(events/yr)
(hours/yr)
(% of MWh)
Economics in Average Year
Energy Price
Capacity Price
Supplier Net Revenue
Average Customer Cost
Total Customer Costs
($/MWh)
($/kW‐yr)
($/kW‐yr)
(¢/kWh)
($B/Yr)
$58
$0
$122
10.1¢
$35.7
Economics in Top 10% of Years
Energy Price
Capacity Price
Supplier Net Revenue (Unhedged)
Supplier Net Revenue (80% Hedged)
Average Customer Cost (Unhedged)
Average Customer Cost (80% Hedged)
Total Customer Costs (Unhedged)
Total Customer Costs (80% Hedged)
($/MWh)
($/kW‐yr)
($/kW‐yr)
($/kW‐yr)
(¢/kWh)
(¢/kWh)
($B/Yr)
($B/Yr)
$99
$0
$362
$244
15.1¢
12.6¢
$53.6
$44.7
9.3%‐12.9%
Capacity Market at 1‐in‐10
Base Case
Sensitivity Cases
14.1%
12.6% ‐ 16.1%
0.10
0.23
0.0001%
0.10 ‐ 0.10
0.22 ‐ 0.23
0.00008% ‐ 0.0001%
$58 ‐ $60
$0 ‐ $0
$97 ‐ $122
10.1¢ ‐ 10.7¢
$35.7 ‐ $37.8
$48
$39
$122
10.2¢
$36.1
$46 ‐ $53
$30 ‐ $60
$97 ‐ $122
10.2¢ ‐ 10.8¢
$36.0 ‐ $38.3
$95 ‐ $102
$0 ‐ $0
$173 ‐ $444
$119 ‐ $259
13.4¢ ‐ 23.0¢
9.8¢ ‐ 21.8¢
$37.4 ‐ $81.5
$34.6 ‐ $77.2
$65
$76
$249
$193
12.9¢
11.7¢
$45.7
$41.5
$58 ‐ $77
$30 ‐ $116
$152 ‐ $302
$128 ‐ $289
12.4¢ ‐ 17.9¢
10.2¢ ‐ 17.7¢
$43.9 ‐ $63.3
$36.2 ‐ $62.9
0.33
0.27 ‐ 0.85
0.86
0.68 ‐ 2.37
0.0004% 0.0003% ‐ 0.0013%
Notes: This reported sensitivity range reflects a subset of the sensitivities we examined in this study, reflecting the following cases: Equal 1/15 Chance of 2011 Weather, No Non‐Weather Load Forecast Error, Perfect Energy Price, and Combustion Turbine as the Marginal Technology. Another important difference between the energy-only and capacity market designs is the nature
of the wholesale price volatility facing customers and suppliers. Increasing the system reserve
margin reduces the level of energy price volatility by mitigating the impact of under-forecasting
load or realizing extreme weather. We illustrate this risk mitigation effect by comparing costs
during the top 10% of all years, reflecting a once-per-decade scarcity year. A customer with 80%
of energy purchases hedged on a seasonal basis and with no capacity hedges would realize onceper-decade costs of 12.6 ¢/kWh (24% above average across all years) under the energy-only
market, and 11.7 ¢/kWh (16% above average) under a capacity market with a 14.1% reserve
margin.2
For suppliers, once-per-decade events affect overall economics even more than for customers,
since increases in energy prices have a proportionally greater impact on suppliers’ energy
margins (since margins derive from the difference between revenues and costs that do not
2
Without hedging, the once-per-decade scarcity year would produce spot prices of 15.1 ¢/kWh (about
50% higher than average costs) under the energy-only market or 12.9 ¢/kWh (26% above average)
under the capacity market at 0.1 LOLE. Actual rates could vary more than these equilibrium
estimates suggest, in part because of changes in gas prices, transmission and distribution rates, and
other factors we did not vary in our analysis.
ix | brattle.com
change). If a supplier is 80% hedged, the once-per-decade year would produce net revenues of
$244/kW-year (2 times CONE) in the energy-only market. In our Base Case simulations with a
mandated 14.1% reserve margin, once-per-decade net revenues (including capacity payments)
are reduced to $193/kW-year (1.6 times CONE). Hence, mandating a reserve margin would
reduce supplier risk. It could also slightly lower their cost of capital and thus the level of longterm average prices, although we have neither estimated nor accounted for this effect.
As noted above, we estimate that ERCOT’s current energy-only market design will support
average reserve margins of approximately 11.5%, slightly above our 10.2% risk-neutral economic
optimum estimate. However, policymakers and stakeholders may still wish to implement a
capacity market to achieve its risk mitigation, reliability, and other potential benefits. For
example, a higher mandated reserve margin may be justified if policymakers and stakeholders
place a greater weight on potential high-cost and low-reliability outcomes that can result from
extreme weather, unexpectedly high load growth, unusual generation outages, or modelling
uncertainties.
This study provides informative reference points but does not constitute a full cost-benefit
analysis of energy-only and capacity market designs. A full cost-benefit analysis would need to
include an explicit consideration of other quantitative and qualitative factors including: (a) cost
reductions from eliminating current demand response programs and moving those resources into
the capacity market; (b) potential value from a capacity market that better coordinates the timing
of investment decisions, which could avoid some boom-bust effects and narrow the distribution
of reserve margin outcomes, thereby further reducing the frequency and costs of shortage events;
and (c) implementation, overhead, and software costs associated with moving to a capacity
market design.
In considering mandating a reserve margin and implementing a capacity market, the Commission
would also have to evaluate many practical implications. Mandating a reserve margin and
establishing a capacity market that can efficiently meet that requirement would require
extensive stakeholder discussions about alternative market design elements such as: (1) the
resource adequacy requirement itself; (2) the implementation of that requirement in a capacity
market, which could involve administratively defining a sloped demand curve; (3) the forward
period and rules for forward and incremental auctions; (4) how to represent transmission
constraints; (5) participation and verification rules for all types of resources; (6) definition of
penalties and performance incentives; (7) market monitoring rules; and (8) settlement processes
for both suppliers and load-serving entities. Although design decisions can benefit from other
regions’ past decade of experience with capacity markets, the process could take two years and
would likely be contentious. In addition, initial design elements would need to be revisited and
likely re-litigated over time as market conditions and reliability challenges change.
Some market participants may also suggest that implementing a capacity market would reduce
the need to maintain a high energy-market price cap or administrative scarcity pricing
mechanisms. We urge against this line of reasoning since efficient energy and ancillary services
prices are important for maintaining operational efficiency and reliability during scarcity periods
x | brattle.com
and during a large variety of operational challenges that resource adequacy does not address.
Maintaining a price cap equal to the value of lost load (VOLL) during outages and prices
reflective of marginal system costs in other types of scarcity events will provide efficient signals
necessary for market-based responses from generators and demand response. These benefits of
efficient energy and ancillary services pricing make them an essential component of the market
design, whether or not a resource adequacy requirement is adopted for ERCOT.
We recognize that concurrent with the completion of our study, ERCOT has released an updated
load forecast that we have not had time to incorporate and consider in our analyses. Because the
updated forecast is much lower than prior forecasts, this may reduce the real or perceived
urgency of addressing the resource adequacy question. In terms of the impact on our study
results, we do not expect that the new forecast would change our estimates of optimal or
equilibrium reserve margins substantially, since those are expressed as a percentage of peak load.
However, our results could change if the new forecast methodology produces very different
distributions of weather and non-weather forecast errors, or if it accounts for demand response
very differently.
Regarding the urgency of the issue, that is a judgment for the Commission, ERCOT, and
stakeholders. The Commission could decide not to act now or, if they see a need to change the
market design in the long term to meet policy objectives, they could take advantage of the
current slack to implement changes with less risk of transitional rate shocks. In any case,
providing stakeholders a clear understanding of whether, how, and when the market design will
change would reduce regulatory uncertainty and benefit market participants as they make
business decisions over the coming years.
xi | brattle.com
I.
Motivation and Context
We have been asked by the Public Utility Commission of Texas (PUCT) and the Electric
Reliability Council of Texas (ERCOT) to estimate the economically optimal reserve margin for
ERCOT’s wholesale electric market. This study contributes to an already substantial body of
technical work, regulatory proceedings, and market design revisions related to the policy
framework for ERCOT’s market design for resource adequacy.3
As we explained in our June 2012 study, the primary resource adequacy challenge in Texas is that
the energy market is likely to support a reserve margin below what is required to meet the
traditional resource adequacy target, which is based on achieving a 1-event-in-ten-years (1-in10) loss of load event (LOLE) standard.4 In most of North America, the planning reserve margin
required to meet a 1-in-10 standard is achieved through either utility planning or capacity
markets.5 This is unlike ERCOT and other energy-only markets, which have no mechanism for
ensuring a particular reserve margin. Instead, the reserve margin is the result of market forces:
low reserve margins create high energy and ancillary service (A/S) prices sufficient to attract
investments in new generation; those investments will continue until high reserve margins result
in prices too low to support any more investment. The resulting equilibrium reserve margin may
be above or below the reserve margin that would be consistent with the 1-in-10 standard,
depending on market fundamentals and uncertainties such as weather.
Our probabilistic market simulations suggest that the equilibrium reserve margin supported by
ERCOT’s energy and A/S markets is below the reserve margin that would yield a 1-in10 LOLE
standard.6 To partially address this gap, ERCOT is implementing a number of market design
changes, including increasing the price cap to $9,000/MWh and introducing a scarcity pricing
mechanism.7 While these changes will improve the efficiency of prices during scarcity events
and increase the equilibrium reserve margin, our prior analyses indicate that these design
changes will not be sufficient to yield a 1-in-10 LOLE outcome.
3
For access to a comprehensive set of associated studies and regulatory proceedings, see the Resource
Adequacy portion of ERCOT’s website as well as the PUCT proceedings under project 40,000. See
ERCOT (2014a) and PUCT (2014).
4
For additional background and information on the LOLE standard, ERCOT’s resource adequacy
challenge, see Newell, et al. (2012).
5
For a more comprehensive review of different regulated and market-based approaches to resource
adequacy, see Spees, et al. (2013), or Pfeifenberger, et al. (2009)
6
See our most recent analysis of the likely equilibrium reserve margin from 2013, Newell, et al. (2013a).
7
ERCOT has implemented a large number of such related changes, but two of the most important are
to increase the price cap to $9,000/MWh and to implement an administrative operating reserves
demand curve (ORDC) for producing high administratively-set prices during scarcity events. See
ERCOT (2013c).
1 | brattle.com
This suggests that ERCOT’s traditional 1-in-10 resource adequacy “target” is inconsistent with
the energy-only market design. Reconciling this incompatibility will require the Commission to
clarify its policy objectives with respect to resource adequacy and adopt a market design that best
supports those objectives. The central market design choices are either: (1) to maintain the
current energy-only market and accept reserve margins that are likely to fall below the current
1-in-10 target; or (2) to impose a mandatory reserve margin standard, likely through a
centralized capacity market, with the reserve margin set at a level consistent with policy
objectives.
The task of our present study is to provide additional information to the PUCT, ERCOT, and
stakeholders as they continue to address these questions. The primary result of our study is an
estimate of the economically optimal reserve margin in ERCOT, defined as the reserve margin
that minimizes total system costs as illustrated in Figure 1. As Figure 1 shows, higher reserve
margins are associated with the higher capital costs of building more capacity (dark blue line).
This higher capital cost is offset by a reduction in the frequency and magnitude of costly
reliability events (light blue line), such as load shed events, emergency events, demand-response
curtailments, and the dispatch of high-cost resources. The tradeoff between increasing capital
costs and decreasing reliability-related operating costs results in a U-shaped total costs curve (red
line), with costs minimized at what we refer to as the “economically optimal” reserve margin.8
8
In developing our approach to calculating the economically optimal reserve margin, we draw upon a
large body of prior work conducted by ourselves and others, although the majority or all of this prior
work was relevant in the context of regulated planning rather than restructured markets. For
example, see Poland (1988), p.21; Munasinghe (1988), pp. 5–7, 12–13; and Carden, Pfeifenberger, and
Wintermantel (2011).
2 | brattle.com
Expected System Cost
Figure 1 “Economically Optimal” Reserve Margin at Lowest System Cost (Illustrative Schematic, Not Simulation Results) Total Cost Lowest
System
Cost
Capacity Capital Costs
VOLL, Emergency Events, and Production Costs 0% 10% Optimal Reserve 20%
Margin Reserve Margin (%)
We conducted reliability and economic market simulations to estimate this economically optimal
reserve margin in ERCOT, and to better inform the economic and reliability implications of
addressing the policy questions at hand. In particular, we examined: (a) the uncertainty range of
both reliability-based and economically-based reserve margin estimates; (b) the system cost and
risk implications of imposing a reserve margin requirement above the level that ERCOT’s
energy-only market would support; (c) the energy and capacity market prices and uncertainties
that would result from mandating different reserve margin levels; and (d) the implications of
different levels of mandated reserve margins on supplier net revenues and total customer costs.
II. Study Assumptions and Approach
In this Section, we provide a summary of the market simulations that we use to analyze the
reliability and economic implications of varying the system reserve margin and generate the
results presented in Sections III and IV. Our simulation modeling approach relies on a detailed
representation of the ERCOT system, including: load and weather uncertainties; the cost and
performance characteristics of ERCOT’s generation and demand-response resources; a
representation of the ERCOT energy and ancillary services market; and a unit commitment and
economic dispatch of all generation resources, demand-response resources, and the transmission
interties with neighboring markets. These simulations are consistent with current projections
for calendar year 2016 in terms of generation fleet, demand-response penetration, fuel prices,
and energy market design.
3 | brattle.com
A. SERVM PROBABILISTIC MODELING FRAMEWORK
We use the Strategic Energy Risk Valuation Model (SERVM) to estimate the economically
optimal reserve margin in the ERCOT system.9 Like other reliability modeling tools, SERVM
probabilistically evaluates resource adequacy conditions by simulating generation availability,
load profiles, load uncertainty, inter-regional transmission availability, and other factors. Based
on these reliability simulations, SERVM estimates standard reliability metrics including loss of
load events (LOLE), loss of load hours (LOLH), and expected unserved energy (EUE). Unlike
other reliability modeling packages, however, SERVM also simulates economic outcomes
including hourly generation dispatch, import-export dynamics, ancillary services, and individual
emergency procedures. SERVM estimates hourly and annual production costs, customer costs,
market prices, net import costs, load shed costs, and generator energy margins as a function of
the planning reserve margin. These results allow us to compare these variable costs against the
incremental capital costs required to achieve higher planning reserve margins.10
The multi-area economic and reliability simulations in SERVM include an hourly chronological
economic dispatch that is subject to inter-regional transmission constraints. Each generation
unit is modeled individually, characterized by its own economic and physical characteristics.
Planned outages are scheduled in off-peak seasons to minimize the impact on reliability, while
unplanned outages and derates occur probabilistically using historical distributions of time
between failures and time to repair, as explained in Section II.C. Load, hydro, wind, and solar
conditions are modeled based on profiles consistent with individual historical weather years.
Dispatch limitations and limitations on annual energy output are also imposed on certain types of
resources such as demand response, hydro generation, and seasonally mothballed units.
The model implements a week-ahead and then multi-hour-ahead unit commitment algorithm
considering the outlook for weather and planned generation outages. In the operating day, the
model runs an hourly economic dispatch of baseload, intermediate, and peaking resources,
including an optimization of transmission-constrained inter-regional power flows to minimize
total cost. During most hours, hourly prices reflect marginal production costs, with higher prices
being realized when import constraints are binding. During emergency and other peaking
conditions, SERVM simulates scarcity prices that exceed generators’ marginal production costs as
explained further in Section II.F below.
9
SERVM software is a product of Astrape Consulting, co-authors of this report, see Astrape (2014)
10
Note that SERVM as a modeling tool does not endogenously estimate capital costs, which are reflected
as a fixed annual cost (in $/kW-year). Total capacity costs simply increase as a function of the
planning reserve margin. The question of whether a particular reserve margin is actually achievable
or realistic under various market designs, including in regions with capacity markets or energy-only
markets, depends on whether those markets have been constructed such that investors are able to
recover their fixed costs at that reserve margin.
4 | brattle.com
To examine a full range of potential economic and reliability outcomes, we implement a Monte
Carlo analysis over a large number of scenarios with varying demand and supply conditions.
Because reliability events occur only when system conditions reflect unusually high loads or
limited supply, these simulations must capture wide distributions of possible weather, load
growth, and generation performance scenarios. This study incorporates 15 weather years,
5 economic load forecast error points, and 100 draws of generating unit performance for a total of
7,500 iterations for each simulated reserve margin case, with each iteration simulating 8,760
hours for the year 2016.11 The large number of simulations is necessary to accurately
characterize from a probabilistic perspective the reliability and economic implications of varying
reserve margins. A probabilistic approach is needed because the majority of reliability-related
costs are associated with infrequent and sometimes extreme scarcity events. Such reliability
events are typically triggered by rare circumstances that reflect a combination of extreme
weather-related loads, high load-growth forecast error, and unusual combinations of generation
outages.
To properly capture the magnitude and impact of reliability conditions during extreme events, a
critical aspect of this modeling effort is the correct economic and operational characterization of
emergency procedures. For this reason, SERVM simulates a range of emergency procedures,
accounting for energy and call-hour limitations, dispatch prices, operating reserve depletion,
dispatch of economic and emergency demand-response resources, and administrative scarcity
pricing.12
B. LOAD MODELING
We model load as an hourly shape across the simulated year, reflecting load diversity between
ERCOT and the neighboring systems. We model weather uncertainty using 15 different load
shapes, consistent with 15 historical weather years and associated hydro, wind, and solar profiles
as explained in Section II.C below. We also model non-weather-related load forecast error
(LFE), which increase with the forward planning period. This non-weather load forecast error,
which has not been reflected in prior reliability analyses of ERCOT’s system, recognizes that, for
example, a three-year-forward reserve margin requirement is associated with more uncertainties
about market conditions during the delivery year than a one-year forward reserve margin
requirement.
11
We implement these SERVM simulations using a planning calendar with only 8760 hours even
though 2016 is a leap year with 8784 hours.
12
Similar to other reliability modeling exercises, our study is focused on resource adequacy as defined by
having sufficient resources to meet peak summer load. As such, we have not attempted to model
other types of outage or reliability issues such as transmission and distribution outages, common mode
failures related to winter weather extremes, or any potential issues related to gas pipeline constraints
or delivery problems.
5 | brattle.com
1. Peak Load and Regional Diversity
Table 1 summarizes the peak load for the ERCOT system and the load diversity relative to the
interconnected neighboring regions. Consistent with the peak load reporting conventions used
in ERCOT’s capacity, demand, and reserves (CDR) report, these peak loads are reported: (a) net
of anticipated load reductions from price-responsive demand (PRD) and load resources (LRs);
and (b) prior to any potential reductions from transmission and distribution service provider
(TDSP) load management or energy efficiency programs.13
Peak load for the ERCOT system was provided by ERCOT staff using a preliminary peak load
estimate as of November 2013, although since that time ERCOT has updated and published a
revised peak load forecast that is somewhat lower than number we use here.14 Hourly load
shapes for ERCOT were developed by ERCOT staff, and are the same load shapes used in the
most recent reliability study of ERCOT’s system from March 2013.15 We independently
developed external regions’ peak load and load shapes based on publicly-available peak load
projections, historical hourly weather profiles, and historical hourly load data.16
The table shows a substantial amount of load diversity between ERCOT and the neighboring
systems, indicating that ERCOT may have access to substantial import quantities during
shortages to the extent that sufficient intertie capability exists. For example, at the time of
ERCOT’s peak load, SPP load is likely to be at only 91% of its own non-coincident peak load,
leaving more than 5,000 MW of excess generation available for export. However, most of these
excess supplies will not be imported because ERCOT is relatively islanded, having only 800 MW
of intertie capability with SPP.
13
See ERCOT (2013a).
14
The new ERCOT Load forecast reports a year 2016 projected peak load of 70,014 MW, or 1,145 MW
lower than the 71,159 MW 50/50 peak load that we use for this study. See ERCOT (2014b).
15
See ECCO (2013a and b).
16
Specifically, SPP and Entergy’s 50/50 peak load forecasts are from the 2012 Long-Term Reliability
Assessment, while the Mexican states’ peak load is based on an assumed 15% reserve margin above the
currently-installed generation fleet, see NERC (2012) and Ventyx (2014). Load shapes in Mexico are
assumed identical to those in ERCOT’s South Zone, as estimated by ERCOT staff; load shapes in SPP
and Entergy are based on our independently-developed statistical relationship between hourly
weather and load estimated over five years of load and weather data from FERC (2013a) and NOAA
(2013).
6 | brattle.com
Table 1 50/50 Peak Loads and Diversity as Used in Reserve Margin Accounting Excluding DR Gross‐Up, Including TDSP Energy Efficiency 50/50 Summer Peak Load
Non‐Coincident
(MW)
Coincident
(MW)
At ERCOT Peak
(MW)
Load Diversity
At Coincident Peak
At ERCOT Peak
(%)
(%)
Reserve Margin at Criterion
At Non‐Coincident Peak
(%)
At ERCOT Peak
(%)
ERCOT
SPP
Mexico
Entergy
Total
71,159
70,106
71,159
56,781
56,080
51,760
9,910
9,484
9,543
26,535
25,766
24,959
164,385
161,436
157,421
1.48%
0.00%
1.23%
8.84%
4.30%
3.70%
2.90%
5.94%
1.79%
4.24%
n/a
n/a
13.6%
24.6%
15.0%
19.4%
12.0%
19.1%
n/a
n/a
Sources and Notes: ERCOT load shapes and 50/50 peak load for 2016 provided by ERCOT staff. Table is consistent with peak loads used in reserve margin accounting (excluding any PRD or LR gross‐up, but including TDSP Energy Efficiency Programs), see ERCOT (2013a). SPP and Entergy peak load forecasts, demand‐response capability, and reserve margin requirements from NERC 2012 Long‐Term reliability assessments, see NERC (2012). SPP and Entergy load shapes developed based on statistical relationships from five years of load data (from FERC Form 714) and 15 years of weather data, see FERC (2013a) and NOAA (2013). Mexico load shape and forecast data were unavailable, assumed a representative 15% reserve margin above generation fleet from Ventyx and a load shape from the ERCOT South zone, see Ventyx (2014). 2. Load Shapes and Weather Years
We represent weather uncertainty by modeling 15 weather years from 1998–2012, as
summarized in Figure 2. The left chart shows projected 2016 peak load relative to the weathernormal peak load before and after load gross-ups for PRD and LRs as explained further in Section
II.D. The chart illustrates asymmetry in the distribution of peak loads, with the highest
projected peak load (consistent with 2011 weather) at 6.6% above weather-normalized peak
loads, compared to a peak load in the mildest weather year that is only 5.2% below weathernormalized peak load.
The right chart in Figure 2 shows the 2016 load duration curves for the 250 highest-load hours
across all 15 weather years. The 2016 load duration curve consistent with a repeat of the extreme
and extended hot summer weather in 2011 is shown as the light blue line in the chart. As
shown, the entire load duration curve from 2011 weather is far above all weather years. This
extreme heat was the driver of a number of emergency events and price spikes during the
summer of 2011, which is described by some as a 1-in-100 weather year. As a result, the
probability assigned to a repeat of 2011-type weather is a source of substantial uncertainty and an
important driver of both reliability and economic results. Consistent with other ERCOT
analyses, we assume a probability of 1% to the 2011 weather pattern, but also report sensitivity
7 | brattle.com
results for a 0% probability and equal probability (1/15 chance) that the 2011 weather pattern
might recur.17
Figure 2 ERCOT Peak Load (Left) and Peak Load Duration Curve (Right) by Weather Year 110%
110%
Peak Load after DR Gross‐Up
105%
Peak Load before DR Gross‐Up
100%
Weather‐
Normal
Peak Load
95%
90%
Peak Load (% of 50/50 Peak)
Peak Load (% of 50/50 Peak)
105%
Weather‐Normal Peak Load
100%
2011 Weather
95%
Weather‐Normal Year
90%
85%
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
85%
0
50
100
150
200
250
Hour of Year
Sources and Notes: ERCOT load shapes for 2016 provided by ERCOT staff. Peak loads are shown before and after gross‐up for PRD and LRs (see Section II.D), load duration curves exclude any gross‐ups . 3. Load Forecast Uncertainty and Forward Period
Forward-looking “planning” or “target” reserve margins differ from actually-realized reserve
margins because both realized peak load and actual available resources can differ from their
projected levels. Realized load will differ from projected load for two reasons. First, due to
weather, because weather cannot be exactly predicted and will cause peak load to differ from the
normalized-weather forecast as explained above. Second, because there are uncertainties in
population growth, economic growth, efficiency rates, and other factors. These non-weather
drivers of load forecast errors (LFEs) differ from weather-related LFEs because they increase with
the forward planning period, while weather uncertainties will generally remain constant and be
independent with the forward period.
As shown in the left chart of Figure 3, we assume that a non-weather LFE is normally distributed
with a standard deviation of 0.8% on a 1-year forward basis, increasing by 0.6% with each
17
Based on a 2011 weather probability assumption provided by ERCOT staff.
8 | brattle.com
additional forward year.18 The distribution includes no bias or asymmetry in non-weather LFEs,
unlike the weather-driven LFE in ERCOT, which has more upside than downside uncertainty.
For our purposes, the relevant forward period for characterizing non-weather LFEs is the period
at which investment decisions must be finalized. We assume investment decisions must be
finalized three years prior to delivery, consistent with the approximate construction lead time for
new generation resources. This means that available supply and the expected planning reserve
margin are “locked in” at three years forward, and the realized reserve margin may differ
substantially as both weather and non-weather uncertainties are resolved as the delivery year
approaches. The right-hand chart of Figure 3 shows the five discrete LFE points we model, equal
to 0%, +/-2%, and +/-4% above and below the forecast. The largest errors are the least likely,
consistent with a normal distribution. We also conduct a sensitivity analysis, examining the
implications on economically optimal and reliability-based reserve margins if the forward period
is varied between zero and four years forward.
Figure 3 Non‐Weather Load Forecast Error LFE with Increasing Forward Period 8%
4%
50/50 Load Forecast
0%
50% Interval
‐4%
90% Interval
‐8%
‐12%
45%
Three‐Year Forward LFE 40%
Discrete LFE Points Modeled 35%
Probability (%)
Forecast Error (% of Peak Load)
12%
30%
25%
20%
15%
10%
5%
1
2
3
4
5
6
7
8
Forward Period (Years)
9
10
0%
‐4%
‐2%
0%
2%
4%
Load Forecast Error (% from 50/50 Forecast)
C. GENERATION RESOURCES
We model the economic, availability, ancillary service capability, and dispatch characteristics of
all generation units in the ERCOT fleet, using plant ratings and online status consistent with
ERCOT’s May 2013 CDR report. We describe here the different approaches we use for modeling
conventional generation, private use networks (PUNs), and intermittent wind and solar. We also
describe here the assumed cost and technical specifications of the gas combined cycle reference
technology that we add or subtract to vary the system reserve margin.
18
This assumed LFE outlook is a standard assumption that we developed in lieu of any ERCOT-specific
analysis, which would require either a longer history of load forecasts in ERCOT or a new analysis
developed out of ERCOT’s peak load forecast, neither of which are currently available.
9 | brattle.com
1. Conventional Generation Outages
A major component of reliability analyses is modeling the availability of supply resources after
considering planned and forced outages. We model forced and maintenance outages of
conventional generation units stochastically, with partial and full forced outages occurring
probabilistically based on distributions accounting for time-to-fail, time-to-repair, startup failure
rates, and partial outage derate percentages. Maintenance outages also occur stochastically, but
SERVM accommodates maintenance outages with some flexibility to schedule maintenance
during off-peak hours. Planned outages are differentiated from maintenance outages and are
scheduled prior to each hourly simulation to occur during low demand periods in the spring and
fall, such that the highest coincident planned outages occur in the lowest load days. This outage
modeling approach allows SERVM to recognize some system-wide scheduling flexibility while
also capturing the potential for severe shortages caused by a number of coincident unplanned
outages.19
We develop distributions of outage parameters for time-to-fail, time-to-repair, partial outage
derate percentages, startup probabilities, and startup time-to-repair from historical Generation
Availability Data System (GADS) data for individual units in ERCOT’s fleet, supplemented by
asset class average outage rates provided by ERCOT where unit-specific data were unavailable.
Table 2 summarizes fleet-wide and asset-class outage rates, including both partial and forced
outages.
Table 2 Forced Outage Rates by Asset Class and Fleet Average Unit Type
Equivalent Forced Mean Time Outage Rate
to Fail
(%)
(hours)
Nuclear
Coal
Gas Combined Cycle
Gas Combustion Turbine
Gas Steam Turbine
1.6%
5.8%
5.5%
12.3%
7.8%
Fleet Weighted Average
6.8%
9,352
878
681
285
325
Mean Time to Repair
(hours)
68
37
37
40
27
Sources and Notes: Parameter distributions based on unit‐specific GADS data and asset class average outage rates from ERCOT. 19
Capturing the possibility of such low-probability, high-impact events is an advantage of the unitspecific Monte Carlo outage modeling used in SERVM. Other production cost and reliability models
typically use a simpler convolution method, resulting in a distribution of outages that may underestimate the potential for extreme events, especially in small systems.
10 | brattle.com
2. Private Use Networks
We represent generation from Private Use Networks (PUNs) in ERCOT on a net generation
basis, where the net output increases with the system energy price consistent with historical data
and as summarized in Figure 4. At any given price, the realized net PUN generation has a
probabilistic quantity, with 11 different possible quantities of net generation within each of 15
different price bands.20
Each of the 11 possible quantities has an equal 9.1% chance of
materializing, although Figure 4 reports only the lowest, median, and highest possible quantity.
We developed this probabilistic net PUN supply curve based on aggregate hourly historical net
output data within each of these selected price bands. During scarcity conditions at prices $2,500
or higher, PUN output produces at least 3,900 MW of net generation with an average of 4,700
MW.
We observe a pattern of availability and price-responsiveness consistent with: (a) gross
generation, much of which is fully integrated into ERCOT’s economic dispatch and security
constrained economic dispatch (SCED), resulting in substantial increases in the expected
quantities over moderate price levels, minus (b) gross load, which introduces some probabilistic
uncertainty around net generation, minus (c) some apparent load price-responsiveness, which
likely contributes to some small additional increase in net PUN generation at very high prices.
Figure 4 PUN Net Generation Energy Price ($/MWh)
$2,500
$2,000
Lowest
Quantity
$1,500
50/50 Quantity
Highest
Quantity $1,000
$500
Average of Binned
Historical Data
$0
0
1,000
2,000
3,000
4,000
5,000
6,000
7,000
PUN Net Generation (MW)
Sources and Notes: Hourly net PUN generation from ERCOT, hourly prices from Ventyx (2014). Individual data points represent summary of data in a series of data binned by price level, within each price bin, the points on the chart represent the lowest 9.1%, middle 9.1%, and top 9.1% of realized quantities in calendar year 2011. 20
Hourly net PUN output data gathered from ERCOT, hourly price data from Ventyx (2014).
11 | brattle.com
3. Intermittent Wind and Solar
We model a total quantity of intermittent wind and solar photovoltaic resources from ERCOT’s
May 2013 CDR report, including the installed capacity of all existing and planned resources as of
2016.21 This includes 15,160 MW nameplate capacity of wind and 124 MW nameplate of solar,
with intermittent output based on hourly generation profiles that are specific to each weather
year.
We developed our system-wide hourly wind profiles by aggregating 14 years of hourly wind
shapes for individual units across the system wind shapes over 1998 to 2011, as provided by
ERCOT staff.22 Figure 5 plots the average wind output by month and time of day, showing the
highest output overnight and in winter months with the lowest output in mid-day and in
summer months. The overall capacity factor for wind resources is 36.9%; although we calculate
reserve margins assuming an effective load-carrying capability of 8.7% consistent with the
ERCOT May 2013 CDR convention.23 As in other systems, it is likely that the estimate of wind
ELCC will change over time as system conditions change or accounting conventions change.
Changing this assigned ELCC would change the reserve margins that we report in our study, and
would require only a simple accounting translation to adjust.
21
See ERCOT (2013a).
22
We aggregated unit-specific output profiles for all units, including traditional and coastal units.
ERCOT obtained the original wind profiles through AWS True Power. Because ERCOT did not
provide a profile for the year 2012, we use the 2011 wind profile to reflect 2012 as well. For new
resources without an identified location, we used the fleet-average profile as aggregated from among
existing resources.
23
See ERCOT (2013a), p. 31.
12 | brattle.com
Figure 5 Average Wind Output by Month and Time of Day Average Output (% Nameplate)
60%
50%
Spring
40%
Winter
30%
Fall
20%
Summer
10%
0%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Hour of Day
Sources and Notes: Average of 15 years’ hourly wind profiles provided by ERCOT, originally from AWS True Power. We similarly model hourly solar photovoltaic output based on hourly output profiles that are
specific to each weather year, as aggregated from county-specific output profiles over years 1998
to 2010.24 In aggregate, solar resources have a capacity factor of 25.4% across all years, and we
assign a 100% of nameplate contribution toward the reserve margin consistent with ERCOT’s
CDR accounting convention.25
4. Hydroelectric
We include 521 MW of hydroelectric resources, consistent with ERCOT’s May 2013 CDR
report.26 We characterize hydro resources using five years of hourly data over 2008–2012
provided by ERCOT, and 15 years of monthly data over 1998–2012 from FERC form 923.27 For
each month, SERVM uses two parameters for modeling hydro resources, as summarized in Figure
24
Individual county output profiles were provided by ERCOT. Because ERCOT did not provide profiles
for the years 2011 and 2012, we adopted the average hourly output profile across years for those
weather years.
25
See ERCOT (2013a).
26
See ERCOT (2013a).
27
See FERC (2013b).
13 | brattle.com
6: (1) monthly total energy output, as drawn from historical data consistent with each weather
year; and (2) daily peak shaving capability, as estimated from historical hourly data.28
When developing hydro output profiles, SERVM will first schedule output up to the monthly
peak shaving capability into the peak hours, but will schedule some output across all hours based
on historically observed output during off-peak periods up to the total monthly output. Hydro
capacity that is not generating power will contribute toward system reserves and count toward
the Operating Reserve Demand Curve (ORDC) x-axis, as described in Section II.F.5.29
Figure 6 Hydro Peak Shaving Capability (left) and Hydro Annual Energy (right) 1,200
1,000
400
300
200
800
600
400
100
200
0
0
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
Annual Energy (GWh)
500
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
Peak Shaving Capability (MW)
600
Sources and Notes:
Monthly and annual energy data from FERC (2013b), peak shaving capability based on five years of historical hourly data from ERCOT. 5. Marginal Resource Technology
To simulate ERCOT’s system across different reserve margins, we must vary the quantity of
installed generating capacity. We vary the quantity of gas combined cycle (CC) plants in our
Base Case, and examine a sensitivity case where we vary the quantity of gas combustion turbines
(CTs). The choice of a gas CC as the reference technology is a change from the assumptions we
28
For years prior to 2008, we did not have hourly hydro data and instead estimated daily peak shaving
capability based on a linear relationship between the two parameters.
29
This remaining hydro capacity is approximately equal to 240 MW on average, consistent with the
contribution that hydrosynchronous resources contribute to ERCOT’s physical responsive capacity
metric as calculated according to ERCOT (2013d), Section 6.5.7.5.
14 | brattle.com
used in our 2012 study, but is consistent with more recent developer announcements that
indicate more CCs will be built than CTs.30
The costs and performance characteristics of the reference CC and CT are summarized in Table 3
and Table 4 respectively. We use the same GE 7FA technology as assumed in our 2012 study,
although we have updated the cost of new entry (CONE) for escalation and cost of capital.31 We
apply 4.3% and 5.2% in cost escalation for the CT and CC respectively, consistent with a oneyear delay in online date from 2015 to 2016.32 We have also updated our estimate of the after-tax
weighted-average cost of capital (ATWACC) for a merchant developer to 8.0% as summarized in
Table 3. This results in an estimated CONE of $97,000/MW-yr and $122,100/MW-yr for the gas
CT and CC respectively, although we also examine a sensitivity range of -10%/+25% in CONE.
Table 3 Gross Cost of New Entry ATWACC
(%/yr)
Gross CONE
Simple Cycle Combined Cycle
($/MW‐yr)
($/MW‐yr)
From 2012 Study (2015 Online Date)
Low: Merchant ATWACC
Mid: ERCOT Planning Assumption
High: Developer‐Reported
7.6%
9.6%
11.0%
$90,100
$105,000
$131,000
$112,400
$131,000
$145,000
Updated Estimate (2016 Online Date)
Low: Base minus 10%
Base: Merchant ATWACC
High: Base plus 25%
n/a
8.0%
n/a
$87,300
$97,000
$121,300
$109,900
$122,100
$152,600
Sources and Notes: 2012 Study numbers and current numbers are adapted from a CONE study for PJM, with adjustments applied as relevant for ERCOT, see Newell, et al. (2012) and Spees, et al. (2011). Updated estimate applies 4.3% and 5.2% escalation derived from Newell, et al. (2013b). 30
For example, Panda Power currently has three combined cycle stations under construction, the Lower
Colorado River Authority’s Ferguson Replacement project will be a combined cycle facility, and FGE
Power recently announced plans to build a combined cycle project in Mitchel County.
31
See Newell, et al. (2012), derived from CONE numbers originally from Spees, et al. (2011).
32
These escalation rates are derived from a study of component-specific escalation rates as documented
in Newell, et al. (2013b).
15 | brattle.com
Table 4 Performance Characteristics Simple Cycle
Plant Configuration
Turbine
Configuration
GE 7FA.05 GE 7FA.05
2 x 0
2 x 1
Heat Rate (HHV)
Base Load
(btu/kWh)
Non‐Summer
(btu/kWh)
Summer
Max Load w/ Duct Firing
(btu/kWh)
Non‐Summer
(btu/kWh)
Summer
Installed Capacity
Base Load
Non‐Summer
Summer
Max Load
Non‐Summer
Summer
Combined Cycle
10,094
10,320
6,722
6,883
n/a
n/a
6,914
7,096
(MW)
(MW)
418
390
627
584
(MW)
(MW)
n/a
n/a
701
656
Sources and Notes: Technical and performance parameters from Spees, et al. (2011). 6. Fuel Prices
We estimate monthly fuel prices for gas, coal, and oil-fired plants in ERCOT and the neighboring
regions using futures prices for the year 2016 and after applying a delivered fuel price basis. We
use Houston Ship Channel, U.S. Gulf Coast, and Powder River Basin as the market price points
for historical and futures prices as shown in Figure 7.33 To estimate a delivered fuel price basis
for each market, we calculated the historical difference between that market price point and
prices as delivered to plants in that region and then escalated the delivered price basis with
inflation to the year 2016.34 This locational basis is inclusive of both market price basis as well as
a delivery charge and therefore may be positive or negative overall as shown in Table 5.
33
Oil futures at WTI Cushing were used to escalate No. 2 fuel oil prices into the future due to lack of
data on No. 2 futures at U.S. Gulf Coast. Data obtained from Bloomberg (2014), and SNL Energy
(2014).
34
Fuel price basis varies by region by not among individual plants. Historical delivered fuel prices from
Ventyx (2014).
16 | brattle.com
Figure 7 Historical and Futures Prices for Gas, Coal, and No. 2 Distillate No 2. Fuel at US Gulf Coast
History Future
$25
Fuel Price ($/MMBtu)
$20
No. 2 Prices Escalated with Oil Futures at WTI Cushing
$15
$10
Gas at Houston Ship Channel HSC Futures
$5
Coal at Powder River Basin
PRB Futures
$0
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
Sources and Notes: No. 2 prices escalated using a linear relationship with WTI Cushing and escalated with WTI futures. Historical and futures prices from Bloomberg (2014) and SNL Energy (2014). Table 5 Delivered Fuel Price Basis including Market Basis and Delivery Charges ERCOT SPP
Entergy
Coal above Powder River Basin
(2016$/MMBtu)
Gas
above Houston Ship Channel
(2016$/MMBtu)
No. 2 Diesel
above U.S.
Gulf Coast
(2016$/MMBtu)
$1.10
$0.93
$1.24
$0.62
$0.90
$0.81
$1.35
$0.84
‐$1.61
Sources and Notes: Locational basis estimated based on delivered fuel prices relative to market price point. Historical delivered prices from Ventyx (2014). D. DEMAND-SIDE RESOURCES
We account for all of the several types of demand resources in ERCOT, including explicitly
modeling how they participate in the energy and ancillary services markets, whether they are
triggered by price-based or emergency actions, and restrictions on availability and call hours.
17 | brattle.com
1. Costs and Modeling of Demand Resources
A number of different types of demand-side resources contribute to resource adequacy in
ERCOT. Table 6 summarizes these resources, explaining how we model their characteristics,
their assumed marginal costs when interrupted, and how they are accounted for in the reserve
margin. We developed these assumptions in close coordination with the ERCOT staff, who
provided assumptions regarding the appropriate losses gross up, and the anticipated quantities of
all types of demand response other than price responsive demand (PRD).
The marginal costs of these demand-side resources are highly uncertain, although the marginal
costs we report in the table are in the general range that we would anticipate given the sparse
data availability. Most of these resources including TDSP load management, emergency response
service (ERS), and load resources (LRs) are dispatched for energy based on an emergency trigger
rather than a price-based trigger consistent with marginal cost. We make the simplifying
assumption that these resources are triggered in order of ascending marginal cost, and at the time
when market prices are equal to their marginal curtailment cost in our Perfect Energy Price Case
as explained further in Section II.F.5 below.
Two types of demand-side resources, TDSP energy efficiency (EE) and self-curtailment to avoid
four coincident peak (4 CP) transmission charges are not explicitly modeled because these load
reductions are already excluded from load shapes and would be realized equally across all
modeled reserve margins. However, these resources are appropriately accounted for using the
conventions of ERCOT’s CDR report as explained further in section II.G below.
18 | brattle.com
Table 6 Summary of Demand Resource Characteristics and Modeling Approach Resource Type Quantity Including Losses Modeling Marginal Cost Approach Adjustments to ERCOT Load Shape Reserve Margin Accounting TDSP Programs Energy Efficiency 553 Not explicitly modeled. n/a None (ERCOT load shapes already reduced for TDSP EE Programs). Load reduction. 541 MW pre‐losses. Load Management 251 Emergency trigger at EEA Level 1. $2,450 None (ERCOT load shapes estimated assuming no LM curtailments). Load reduction. 240 MW pre‐losses. Emergency Response Service (ERS) 10‐Minute ERS 363 MW at Peak. Emergency trigger at EEA Level 2. $3,681 None (ERCOT load shapes estimated assuming no ERS curtailments). Load reduction. 347 MW pre‐losses. Emergency trigger at EEA Level 1. $1,405 None (ERCOT load shapes estimated assuming no ERS curtailments). Load reduction. 77 MW pre‐losses. Grossed up by as much as 195 MW. 1,205 MW load reduction. 496 MW Maximum. 30‐Minute ERS 80 MW at Peak. 133 MW Maximum. Load Resources (LRs) Non‐
Controllable LRs 1,205 MW at Peak. Controllable LRs 36 1,400 MW Maximum. Economically dispatch for RRS (most hours) or energy (few peak hours). Emergency deployment at EEA Level 2. $2,569 Self‐schedule for Regulation in all hours. n/a Remaining 195 MW will be excluded from reported peak load. None. Supply Resource. None (ERCOT load shapes already reduced for 4CP response). None. Voluntary Self‐Curtailments 4 CP Reductions 715 Price Responsive Demand 706 Not explicitly modeled (assume 4CP behavior will persist in all circumstances). n/a Economic self‐
curtailment, but with uncertain availability. $250 ‐ $9,000/MWh Grossed Up. Already excluded from reported peak load. None. Already excluded from reported peak load. Sources and Notes: Developed based on analyses of recent DR participation in each program and input and data from ERCOT staff. See following sections. For 10‐Minute ERS and 30‐Minute ERS there is an 8‐hour call limit per Contract Period. See Table 7 below. TDSP Load Management Programs have a 16‐hour call limit from June to September. 19 | brattle.com
2. Emergency Response Service
Emergency Response Service (ERS) includes two types of products, 10-minute and 30-minute
ERS, with the quantity of each product available changing by time of day and season as shown in
Table 7. The quantity of each product by time of day and season is proportional to the quantities
most recently procured over the four seasons of year 2013, with the 2016 summer peak quantity
assumption provided by ERCOT.35 We apply a losses gross-up to these ERS quantities for
modeling purposes, but do not apply that gross-up when calculating the reserve margin as is the
convention in the CDR report. Demand resources enrolled under ERS are dispatchable by
ERCOT during emergencies, but cannot be called outside their contracted hours and cannot be
called for more than eight hours total per season.36
Table 7 Assumed ERS Quantities Available in 2016 Contract Period
June ‐ September
BH1: Weekdays 8 AM ‐ 1 PM BH2: Weekdays 1 PM ‐ 4 PM
BH3: Weekdays 4 PM ‐ 8 PM*
NBH: All Other Hours
October ‐ January
BH1: Weekdays 8 AM ‐ 1 PM BH2: Weekdays 1 PM ‐ 4 PM
BH3: Weekdays 4 PM ‐ 8 PM
NBH: All Other Hours
February ‐ May
BH1: Weekdays 8 AM ‐ 1 PM BH2: Weekdays 1 PM ‐ 4 PM
BH3: Weekdays 4 PM ‐ 8 PM
NBH: All Other Hours
Summer Peak (June‐Sept, BH3)
Maximum at Any Time
Quantity (w/o Losses)
10‐Min
30‐Min
(MW)
(MW)
Quantity (w/ Losses)
10‐Min 30‐Min Total
(MW)
(MW)
(MW)
475
353
347
387
128
88
77
101
496
369
363
404
133
92
80
106
629
461
443
510
453
450
437
405
81
83
82
72
472
470
456
422
84
86
86
75
557
556
542
497
460
459
443
386
73
73
59
41
480
479
463
403
77
77
62
43
557
556
525
445
347
475
77
128
363
496
80
133
443
629
Sources and Notes: Total available ERS MW for 2016 June‐Sept. BH3 provided by ERCOT staff. ERS 10‐min and 30‐min MW for other contract periods scaled proportionally to summer quantity, based on availability over calendar year 2013, from ERCOT (2013e). ERS resources have an eight‐hour call limit applies to both product types and are not callable outside contracted hours, see ERCOT (2013f‐g). 35
For total ERS procurement quantities by product type and season, see ERCOT (2013e).
36
See ERCOT (2013f–g).
20 | brattle.com
3. Non-Controllable Load Resources
We model 1,400 MW of non-controllable load resources (LRs) that actively participate in the
responsive reserve service (RRS) market.37 These non-controllable LRs are separated into two
dispatch blocks with different costs and dispatch characteristics as summarized in Table 8 and
Figure 8. The larger 1,205 MW block of LRs self-schedule to sell RRS in all hours, but in
emergency conditions ERCOT will dispatch these resources for energy during emergency
conditions as explained further in Section II.F.2.
The smaller 195 MW block of resources is modeled similarly to generation that is dispatched
against an energy “strike price” of $380/MWh. These resources will be scheduled for RRS service
in most hours when energy prices are relatively low; but during peaking events they will act like
demand response that self-curtails to avoid paying those high prices. The approximate
$380/MWh strike price for these resources is based on an analysis of the realized quantities of
non-controllable LRs that withdrew from the RRS market during times of high energy prices as
shown in Figure 8. This modeling approach mimics typical market results in which the vast
majority of hours will attract the maximum 1,400 MW of LRs in RRS, but summer peak hours
will clear a lower quantity.38
Table 8 Non‐Controllable LRs Block Characteristics Block Size
Cumulative Quantity
Marginal Cost
Energy Strike Price
(MW)
(MW)
($/MWh)
($/MWh)
195
1,205
195
1,400
$380
$2,569
$380
$9,000
37
Currently, 1,400 MW is the maximum quantity of non-controllable LRs that are allowed to sell
responsive reserve service (RRS) and is the clearing quantity in the vast majority of hours.
38
Note that the self-curtailed LRs also contribute to resource adequacy by avoiding energy consumption
during peak times similar to PRD, but also like PRD these loads are not included in ERCOT’s load
forecast and so are not an explicit line item in the CDR report.
21 | brattle.com
Figure 8 Marginal Cost of Non‐Controllable Load Resources $3,000
ERCOT Emergency Curtailment Energy Price ($/MWh)
$2,500
$2,000
2011
$1,500
$1,000
2012‐13
$500
Self‐Curtailment $0
0
100
200
300
400
LR Quantity (MW)
Sources and Notes: Non‐controllable LR Quantity on the x‐axis is the quantity not clearing for RRS, and assumed to self‐curtail for energy. LR cleared and uncleared quantities from ERCOT, hourly energy prices from Ventyx (2014). 4. Price-Responsive Demand
To date there has been no comprehensive study of the likely quantity of price responsive demand
(PRD) active in ERCOT, and energy price is not included as a predictive variable in ERCOT’s
load forecast. The quantity of PRD that currently exists is uncertain and likely to grow over the
coming years as the energy market price cap increases to $9,000/MWh and as other scarcity
pricing reforms result in more frequent high price events.
We estimate the approximate quantity of PRD in ERCOT by comparing realized load shapes over
the top 50 hours in the 2008–2011 period to those forecast in ERCOT’s load shapes over the same
weather years. We observe that ERCOT’s forecast of its top 50-hour load shape is “peakier” than
the actual realized load shapes, likely because ERCOT’s predicted load shape is not flattened by
the PRD that responded to high prices. Figure 9 shows the comparison of these two peak load
duration curves for the year 2011, with actual loads grossed up for curtailments in noncontrollable LRs and ERS deployments. Overall, this peakier forecast load shape indicates an
approximate 1% PRD penetration, although with substantial uncertainty.
To model PRD, we first stretch the ERCOT load duration curve in the super-peak hours by up to
1% above the forecast, while maintaining the same load shape and annual energy. We then
model PRD as a supply-side resource, with some uncertainty in the quantity that will be realized
as shown in Figure 10. At the price cap of $9,000/MWh, we assume a 10%, 15%, 50%, and 25%
chance of 0%, 0.5%, 1%, and 1.5% PRD respectively. We calculate the quantity of PRD available
at price levels below the cap assuming a constant elasticity of demand. For comparison in the
22 | brattle.com
chart, we also show our approximate PRD supply curve compared to prices and approximate
PRD quantities as implied by a comparison of predicted and actual load shapes. We stress that
this analysis is a relatively rough approximation of PRD in ERCOT, and would recommend a
more thorough statistical analysis of this resource, including possibly incorporating price as an
explicit variable in ERCOT’s various load modeling exercises.
Load Duration Curve (% of Average Load)
Figure 9 Peak Load Duration Curve 2011 Weather Projected Compared to Actual 185%
Predicted
180%
175%
Actual
170%
165%
0
20
40
Hour of Year 60
80
Sources and Notes: Actual and modeled hourly load data provided by ERCOT. Figure 10 Price Responsive Demand Energy Supply Curve, Compared to Historical $3,000
10% 15% 50% 25% Probability of Drawing this Quantity
RT Prices ($/MWh)
$2,500
$2,000
2011
2008
2012
$1,500
2009
$1,000
2010
$500
$0
0.0%
0.5%
1.0%
1.5%
2.0%
Predicted ‐ Actual Load (% of Peak Load)
2.5%
Sources and Notes: Actual and modeled hourly load data provided by ERCOT, actual prices from Ventyx. (2014). E. ERCOT AND EXTERNAL SYSTEMS’ RESOURCE OVERVIEW
This section of our report provides an overview of the ERCOT fleet as summarized for individual
resource types in Sections II.C and II.D above. We describe the system interconnection
23 | brattle.com
topology, intertie availability, ERCOT and neighboring regions’ resource mixes and supply
curves.
1. System Topology and Intertie Availability
We model ERCOT and the neighboring interconnected regions with the interconnection
topology as shown in Figure 11. ERCOT is a relatively islanded system with only 1,090 MW of
high voltage direct current (HVDC) interties; the majority of that intertie capacity is with SPP.
As explained in Section II.A above, SERVM runs a multi-area economic dispatch and will
schedule imports or exports from ERCOT depending on the relative cost of production compared
to the neighboring systems. During peaking conditions, ERCOT will generally import power due
to the high internal prices, unless imports cannot be realized. ERCOT may not be able to import
during peak conditions because either: (a) the neighboring system experiences a simultaneous
shortage and will prioritize meeting its own load, or (b) insufficient intertie capability exists to
support the desired imports.
Figure 11 System Topology and Modeled Interties SPP 810 MW 5,180 MW
Entergy ERCOT (MISO) 280 MW Mexico (Coahuila, Nuevo Leon, & Tamaulipas)
Sources and Notes: ERCOT intertie ratings from ERCOT staff, SPP‐Entergy path rating from OATI (2013). We model transmission availability according to a probability distribution that is independent of
weather and other variables. Figure 12 shows the cumulative probability distribution of
available transfer capability (ATC), derived from hourly ATC data from 2010–2013 across
ERCOT’s SPP and Mexico interfaces. The SPP-Entergy interface is similarly modeled using a
24 | brattle.com
probability distribution with 95% availability on average. Even if transmission is available,
ERCOT may not be able to import in emergency if the external region is peaking at the same
time.
Figure 12 ERCOT Intertie Capability Cumulative Distribution Function ERCOT Simultaneous
Import Capability
1,200
Import Capability (MW)
1,000
ERCOT↔ SPP
800
600
400
ERCOT↔ Mexico
200
0
0%
10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Percent of Hours
Sources and Notes: Based on hourly historical ATC data provided by ERCOT over 2010‐13. Export capability assumed equal to import capability in any individual draw, which is usually (but not always) the case in actual operations. 2. Resource Mix Across Reserve Margins
Figure 13 summarizes the supply resource mix that we model in ERCOT, SPP, Entergy, and
Mexico. As explained above, we use ERCOT’s May 2013 CDR Report as the authoritative source
for documenting the ERCOT fleet, although we use the following assumptions for some special
resource types:

Switchable Units are included as internal resources, with the 317 MW unit that is
committed off-system excluded from our model;

Retirements are excluded starting in the CDR-specified year;

Seasonal Mothballs are included in our model, with 1,876 MW of seasonal
mothballs that are available for dispatch only in summer months, nearly all of
them over May through September;

Permanent Mothballs are excluded from our model; and

Planned New Units are excluded.
25 | brattle.com
Starting with this baseline of ERCOT resources, we conduct simulations over a range of reserve
margins above and below the existing resource base. To reduce the fleet to the minimum reserve
margin, we exclude all planned new resources and mothballed units (as above), and then exclude
a small number of additional existing gas CCs that are similar to the reference technology
described in Section II.C.5 above. To increase the fleet size, we add a single resource type (a gas
CC in the Base Case) across a wide range of potential reserve margins.
Figure 13 Resource Mix for ERCOT and Neighboring Systems 90,000
Load Resources
Storage
Net PUN Generation
Oil
Gas CTs
Gas STs
Gas CCs
Biomass
Coal
Hydro
Photovoltaic
Wind
Nuclear
80,000
Capacity (ICAP MW)
70,000
Peak Load
60,000
50,000
40,000
30,000
20,000
10,000
0
ERCOT
SPP
Entergy
Mexico
Sources and Notes: ERCOT shown at a 10% reserve margin resources reflect the May 2013 CDR Report as explained above, with the exception of adjustments related to demand‐side resources as instructed by ERCOT and including or excluding resources to adjust the reserve margin as explained in Section II.G, see ERCOT (2013a) Wind and solar reported at 8.7% and 100% of nameplate respectively, consistent with CDR, see ERCOT (2013a). External regions’ generation resource mix is from Ventyx (2014), with demand‐response penetration from NERC (2012). Total resources in external regions are adjusted such that external regions are at their target reserve margins of 13.6%, 12%, and 15% for SPP, Entergy, and Mexico respectively while maintaining the same generation resource mix, see NERC (2012). For neighboring regions, we rely on public data sources for the fleet makeup and demandresponse penetrations.39 We model each external region at criterion, meaning that we treat them
exactly at their respective reserve margin targets of 13.6%, 12%, and 15% for SPP, Entergy, and
Mexico respectively.40 Because these regions are currently capacity long, we adjusted their
39
Specifically, we take external regions resource mix from Ventyx (2014) and external regions’ demandresponse penetrations from NERC (2012).
40
See NERC (2012).
26 | brattle.com
resource base downward by removing individual units of different resource types in order to
maintain the current overall resource mix.
3. Supply Curves
To provide more intuition regarding anticipated prices and intertie flows during normal
conditions, we summarize the ERCOT and neighboring regions’ supply curves in Figure 14. The
curve reports energy dispatch costs consistent with year 2016, accounting for unit-specific
heatrates, variable operations and maintenance (VOM) costs, and locational fuel prices from
Section II.C.6. For ERCOT, we gathered unit-specific information representing heatrate curves,
VOM, ancillary service capabilities, ramp rates, startup fuel, non-fuel startup costs, and run-time
restrictions from ERCOT. For external regions, we gathered unit-specific heatrates from public
data sources, supplemented by class-average characteristics similar to those in ERCOT for other
unit characteristics.41
For all thermal resources, we model a relationship between capacity and hourly temperature
which results in increased output from the fleet during colder periods. Each unit is designated a
specific weather station in which the hourly temperature determines the output of the plant for
that hour. By doing this, we ensure the relationships among load, thermal generation, wind, and
solar across the 15 weather years that are simulated.
Overall, the regions have relatively similar supply curves that will tend to reduce the level of
interchange compared to neighboring regions with very different economics. Interchange will
also be limited because of ERCOT’s relatively small quantity of HVDC interties, having only 810
MW of interties with SPP and 280 MW with Mexico.42 However, some factors affecting the
quantity and economic value of interchange include that: (a) SPP has more lower-cost coal that is
somewhat cheaper than ERCOT-internal resources that are dominated by efficient but somewhat
higher-cost gas CCs, which will lead to ERCOT being a net importer, and (b) Mexico has a
substantial proportion of relatively high-cost oil-fired peaking units, which will make such
imports unlikely except at high prices in shortage conditions. Further, the regions experience
some amount of load diversity that will change the relative economics of supply in each region
and lead to inter-regional flows.
41
We gathered heatrates from Ventyx (2014).
42
Based on several years of historical hourly intertie ratings supplied by ERCOT.
27 | brattle.com
Figure 14 2016 System Supply Curves $300
Marginal Cost ($/MWh)
$250
ERCOT
$200
$150
Mexico
$100
Entergy
$50
SPP
$0
0%
20%
40%
60%
80%
Supply (% of Summer Peak Load)
100%
120%
Sources and Notes: ERCOT is shown at 10% reserve margin, with resource mix consistent with May 2013 CDR as explained in Section II.E.2, using unit‐specific heat rates, VOM, and other characteristics obtained from ERCOT. External systems resource mix from Section II.E.2, with resource attributes from Ventyx (2014). Supply curves reflect VOM and fuel costs, with fuel prices from Section II.C.6 above. F. SCARCITY CONDITIONS
Increasing the reserve margin provides benefits primarily by reducing the frequency and severity
of high-cost emergency events. Calculating the economically optimal reserve margin requires a
careful examination of the nature, frequency, and cost of each type of market-based or
administrative emergency action implemented during such events.
1. Market Parameters
We develop a representation of the ERCOT market consistent for 2016 using the parameters
summarized in Table 9, although we note that some of the simulated market design elements are
still under revision. We assume that the administrative Value of Lost Load (VOLL) is equal to
the true market VOLL and equal to the High System-Wide Offer Cap (HCAP) at $9,000/MWh.43
We also conduct a sensitivity analysis for a reasonable range of VOLL.
43
See PUCT (2012).
28 | brattle.com
Consistent with current market rules, we tabulate the Peaker Net Margin (PNM) over the
calendar year and reduce the System-Wide Offer Cap (SWOC) to the Low System-Wide Offer
Cap (LCAP) of $2,000/MWh after the PNM threshold is exceeded.44 However, we stress that this
mechanism will have much less impact on the market than it has previously because the LCAP
now only affects the Power Balance Penalty Curve (PBPC) and suppliers’ offers, but does not
affect the Operating Reserves Demand Curve (ORDC). Therefore, prices will still rise gradually
to the VOLL of $9,000 in shortage conditions even after the PNM threshold is exceeded, thereby
rendering the LCAP far less important. We further explain our implementation of the PBPC and
ORDC in Sections II.F.4 and II.F.5 below.
Table 9 ERCOT Scarcity Pricing Parameters Assumed for 2016 Parameter
Value of Lost Load
High System‐Wide Offer Cap
Low System‐Wide Offer Cap
Peaker Net Margin Threshold
Value
Notes
$9,000/MWh
$9,000/MWh
$2,000/MWh
$291,000/MW‐yr
Administrative and actual
Always applies to ORDC
Applies only to PBPC
3 x CT CONE
Sources and Notes: HCAP, LCAP, and VOLL parameters consistent with scheduled increases by 2016, see PUCT (2012). PNM threshold is set at three times CT CONE consistent with current market rules and our updated CONE estimate from Section II.C.5, but is lower than the $300,000/MW‐yr value applicable for 2013, see PUCT (2012). The offer cap and PNM parameters determine the maximum offer price for small suppliers in
ERCOT’s market under its monitoring and mitigation framework. However, we do not explicitly
model these dynamics and instead assume that suppliers always offer into the market at price
levels reflective of their marginal costs, including commitment costs.
2. Emergency Procedures and Marginal Costs
The benefits of increasing the system reserve margin accrue from reducing the frequency and
severity of costly scarcity events. Therefore, estimating the economically optimal reserve margin
requires careful representation of the nature, trigger order, and marginal costs realized during
each type of scarcity event. Table 10 summarizes our modeling approach and assumptions under
all scarcity and non-scarcity conditions depending on what type of marginal resource or
administrative emergency procedure would be implemented to meet an incremental increase in
demand. These marginal resources are listed in the approximate order of increasing marginal
costs and emergency event scarcity, although in some cases the deployment order overlaps.
We distinguish between market-based responses to high prices in scarcity conditions and out-ofmarket administrative interventions triggered by emergency conditions. Among market-based
44
See PUCT (2012).
29 | brattle.com
responses, we include generation, imports, and price-responsive demand, including some very
high-cost resources that will not economically deploy until prices are quite high. We also model
reserve shortages that are administrative in nature, but triggered on a price basis consistent with
the ORDC and PBPC as explained in the following sections.
A final category of emergency interventions encompasses out-of-market actions including ERS,
LR, TDSP load management, and firm load shed deployments that are triggered for non-price
reasons during emergency conditions. We implement each of these actions at a particular
shortage level as indicated by the quantity of reserves capability available according to the ORDC
x-axis, a measure similar to the physical responsive capacity (PRC) indicator used by ERCOT to
monitor system operations. To estimate the approximate ORDC x-axis at which each action
would be implemented, we reviewed ERCOT’s emergency operating procedures, evaluated the
PRC level coinciding with each action during historical emergency events, and vetted these
assumptions with ERCOT staff.45 These trigger levels are in line with historical emergency
events, although actual emergency actions are manually implemented by the system operator
based on a more complex evaluation of system conditions, including frequency and near-term
load forecast.
We also describe here the marginal costs of each type of scarcity event as well as the prevailing
market price during those events. In a perfectly-designed energy market, prices would always be
equal to the marginal cost that would theoretically lead to optimal response to shortage events
and an optimal level of investments in the market. In ERCOT, prices are reflective of marginal
costs in most cases but not all. Specifically, the ORDC curve is designed based on an assumption
that load would be shed at X = 2,000 MW, while our review of historical events indicates that
load shedding is more likely to occur at a lower level of X = 1,150 MW. This discrepancy results
in prices above marginal costs during moderate scarcity events, as discussed further in Section
II.F.5 below.
45
The PRC metric is calculated with some accounting nuances that make it a somewhat different
number from the ORDC Spin x-axis, we do not consider these nuances in our modeling, for the
formula for calculating PRC, see ERCOT (2013d), Section 6.5.7.5.
30 | brattle.com
Table 10 Emergency Procedures and Marginal Costs Emergency Level Marginal Resource Trigger Price Marginal System Cost n/a Generation Price Approximately $20‐$250 Same n/a Imports Price Approximately $20‐$250
Up to $1,000 during load shed Same n/a Non‐Spin Shortage Price Marginal Energy +
Non‐Spin ORDC w/ X = 2,000 Marginal Energy +
Non‐Spin ORDC w/ X = 1,150 n/a Emergency Generation Price $500 Same n/a Price‐Responsive Demand Price $250‐$9000 Same n/a Spin Shortage Price Marginal Energy + Non‐Spin + Spin ORDC w/ X = 2,000 Marginal Energy + Non‐Spin
+ Spin ORDC w/ X = 1,150 n/a Regulation Shortage Price Power Balance Penalty Curve Same
(Unless Capped by LCAP) EEA 1 30‐Minute ERS Spin ORDC x‐axis
= 2,300 MW $3,239 at Summer Peak
(from ORDC) $1,405 EEA 1 TDSP Load Curtailments Spin ORDC x‐axis
= 1,750 MW $9,000 (from ORDC) $2,450 EEA 2 Load Resources in RRS Spin ORDC x‐axis
= 1,700 MW $9,000 (from ORDC) $2,569 EEA 2 10‐Minute ERS Spin ORDC x‐axis
= 1,300 MW $9,000 (from ORDC) $3,681 EEA 3 Load Shed Spin ORDC x‐axis
= 1,150 MW VOLL = $9,000 Same Sources and Notes: Developed based on review of historical emergency event data, input from ERCOT staff, and ERCOT’s emergency procedure manuals; see ERCOT (2013d), Section 6.5.9.4, and ERCOT (2013h), Section 4 3. Emergency Generation
During severe shortage conditions, there are out-of-market instructions by ERCOT as well as
strong economic incentives for suppliers to increase their power output to their emergency
maximum levels for a short period of time.46 During these conditions, suppliers can output
power above their normal capacity ratings, although doing so is costly because it may impose
additional maintenance costs and may put the unit at greater risk of failure.
46
CITE ERCOT procedures asking for emergency output.
31 | brattle.com
To estimate the approximate quantity and cost of emergency generation, we reviewed ERCOT
data on units’ emergency maximum ratings as well as the actual realized output levels during
high-price events in August 2011 as summarized in Figure 15. According to ERCOT’s emergency
maximum ratings, the aggregate ERCOT fleet should be able to produce approximately 360 MW
in excess of summer CDR ratings.47 This is approximately consistent with the quantity of output
we observed historically during high-price events, although the actual realized quantity above
summer CDR ratings is somewhat uncertain. To reflect this uncertainty and range of possible
realized emergency generation, we model emergency generation probabilistically assuming 50%
chance of realizing either 230 MW or 360 MW. We estimate the marginal cost of emergency
output at approximately $500/MWh, consistent with the historical price levels at which we
observed substantial output in excess of summer CDR ratings as shown in Figure 15.
Figure 15 Emergency Generation Above CDR Ratings During High‐Price Events $3,500
RT Price ($/MWh)
$3,000
$2,500
$2,000
$1,500
$1,000
Marginal Cost $500
$0
0
100
200
300
400
Emergency Output (MW)
Sources and Notes: Unit‐specific hourly generation less summer CDR ratings treated as “emergency generation” aggregated on a fleet‐wide basis. Generation data from ERCOT, hourly energy prices from Ventyx (2014). 4. Power Balance Penalty Curve
The Power Balance Penalty Curve (PBPC) is an ERCOT market mechanism that introduces
administrative scarcity pricing during periods of supply shortages. The PBPC is incorporated
47
This number excludes private use network resources, which we model separately as explained in
Section II.C.2 above.
32 | brattle.com
into the security constrained economic dispatch (SCED) software as a set of phantom generators
at administratively-specified price and quantity pairs, as summarized in the blue curve in Figure
16.48 Whenever a PBPC is dispatched for energy, it reflects a shortage of supply relative to
demand in that time period that, if sustained for more than a moment, will materialize as a
reduction in the quantity of regulating up capability. At the highest price, the PBPC will reach
the system-wide offer cap (SWOC), which is set at the HCAP at the beginning of each calendar
year but which will drop to the LCAP if the PNM threshold is exceeded as explained in
Section II.F.1 above.
We similarly model the PBPC as phantom supply that may influence the realized price, and that
will cause a reduction in available regulating reserves whenever called. However, we model only
the first 200 MW of the curve at prices below the cap, and assume that all price points on the
PBPC will increase in approximate proportion to the upcoming scheduled increases in the
SWOC.49 We also assume that the prices in the PBPC are reflective of the marginal cost incurred
by going short of each quantity of regulating reserves.50 Consistent with current market design,
we assume that once the PNM threshold is exceeded, the maximum price in the PBPC will be set
at the LCAP + $1/MWh or $2,001/MWh.51 Note that even after the maximum PBPC price is
reduced, ERCOT market prices may still rise to a maximum value of VOLL equal to $9,000/MWh
during shortages because of the ORDC as explained in the following section.
48
CITE the below ERCOT source explaining PBPC and the
49
Price points in the revised PBPC are approximately in proportion to the scheduled price cap increase
between 2013 and 2016, although the exact prices are set at rounded values based on ERCOT staff
input. Year 2013 PBPC numbers from ERCOT (2013b).
50
Once the PNM is exceeded and the PBPC is reduced, these prices are no longer reflective of marginal
cost but are instead lower than marginal cost at regulation shortage quantities greater than 40 MW.
51
See ERCOT (2013b).
33 | brattle.com
Figure 16 Power Balance Penalty Curve Power Balance Penalty Curve Price ($/MWh)
$10,000
$9,000
PBPC in 2016
Prices Increased in Approximate
Proportion to Price Cap
$8,000
$7,000
$6,000
PBPC in 2013
$5,000
$4,000
$3,000
$2,000
2016 PBPC If PNM Threshold is Exceeded
Prices Capped at LCAP + $1
$1,000
$0
0
50
100
150
200
Power Balance Quantity Violation (MW)
250
Sources and Notes: Year 2016 PBPC updated in approximate proportion to the scheduled increases in system price cap, as rounded up or down consistent with ERCOT staff guidance. 2013 PBPC numbers from ERCOT (2013b), p. 23. 5. Operating Reserves Demand Curve
The most important and influential administrative scarcity pricing mechanism in ERCOT is the
recently-approved operating reserves demand curve (ORDC) that reflects the willingness to pay
for spinning and non-spinning reserves in the real-time market.52 Figure 17 illustrates our
approach to implementing ORDC in our modeling, which is similar to ERCOT’s implementation
although with some simplifications.53 We implement all 48 distinct ORDC curves, with different
curves, four seasons, six times of day, and two types of operating reserves.54
52
Note that the ORDC is not planned to be co-optimized with the energy market at this time, but the
real-time spinning and non-spinning prices they produce are used to settle against the day-ahead RRS
(Spin) and NSRS (Non-Spin) markets.
53
For a detailed explanation of ERCOT’s ORDC implementation see their whitepaper on the
methodology for calculating ORDC at ERCOT (2013c).
54
See ERCOT (2013c), p. 15.
34 | brattle.com
Figure 17 Operating Reserve Demand Curves Example: Summer Hours 15‐18 $9,000
Price or Marginal System Cost ($/MWh)
$8,000
Marginal Cost
Marginal Energy plus
ORDC w/ X = 1,150
RRS Price
Energy Price
Marginal Energy plus
ORDC w/ X = 2,000
$7,000
$6,000
$5,000
NSRS Price
$4,000
Marginal Energy Plus NSRS ORDC
w/ X = 1,150
$3,000
Marginal Energy Plus Non‐Spin ORDC
w/ X = 2,000
$2,000
$1,000
Marginal Energy Cost $0
0
500
1,000
1,500
2,000
2,500
3,000
3,500
ORDC Spin plus Non‐Spin X‐Axis (MW)
Sources and Notes: ORDC curves developed consistent with ERCOT (2013c). 4,000
4,500
The ORDC curves are calculated based on a loss of load probability (LOLP) at each quantity of
reserves remaining on the system, multiplied by the value of lost load (VOLL) caused by running
short of operating reserves.55 This curve reflects the incremental cost imposed by running short
of reserves and is added to the marginal energy cost to estimate the total marginal system cost
and price.
The x-axis of the curve reflects the quantity of operating reserves available at a given time,
where: (a) the spin ORDC includes all resources providing regulation up or RRS, suppliers that
are online but dispatched below their maximum capacity, hydrosynchronous resources, non-
55
Note that the lost load implied by this function and caused by operating reserve shortages is additive
to the lost load that we report elsewhere in this study. This is because the LOLP considered in
ERCOT’s ORDC curve is caused by sub-hourly changes to supply and demand that can cause shortterm shortages and outages that are driven only by small quantities of operating reserves, but are not
caused by an overall resource adequacy shortage, which is the type of shortage we model elsewhere in
this study. For simplicity and clarity, we refer to these reserve-related load-shedding events as
“reserve shortage costs” to distinguish them from the load shedding events caused by total supply
shortages. We do not independently review here ERCOT’s approach to calculating LOLP, but instead
take this function as an accurate representation of the impacts of running short of operating reserves.
35 | brattle.com
controllable load resources, and 10-minute quickstart; and (b) the spin + non-spin ORDC include
all resources contributing to the spin x-axis as well as any resources providing NSRS and all 30minute quickstart units. Table 11 provides a summary of the resources that are always available
to contribute to the ORDC x-axis unless they have been dispatched for energy although the
realized ORDC x-axis can be higher (if other resources are committed but not outputting at their
maximum capability) or lower (during peaking conditions when some of the below resources are
dispatched for energy).56
Table 11 Resources Always Contributing to ORDC X‐Axis Unless Dispatched for Energy Spin X‐Axis
Controllable Load Resources Hydrosynchronous Resources
10‐Minute Quickstart
Non‐Controllable Load Resources
(MW)
(MW)
(MW)
(MW)
36
240
2,283
1,400
Non‐Spin X‐Axis
30‐Minute Quickstart
(MW)
2,053
Total Spin + Non‐Spin (MW)
6,012
The red and pink curves in Figure 17 show the ORDC curves used for price-setting purposes,
calculated as if ERCOT would shed load at an ORDC x-axis of X = 2,000 MW. However, as we
explained in Section II.F.2 above, we assume that load shedding will actually occur at X = 1,150
MW based on our analysis of recent emergency events and consistent with the blue curves
below. In other words, we model a discrepancy between marginal costs (blue) and market prices
(red) that will create some inefficiency in realized market outcomes.
As in ERCOT’s ORDC implementation, we calculate: (a) non-spin prices using the non-spin
ORDC; (b) spin prices as the sum of the non-spin and spin ORDC; and (c) energy prices as the
sum of the marginal energy production cost plus the non-spin and spin ORDC prices. However,
as a simplification we do not scale the ORDC curves in proportion to VOLL minus marginal
energy in each hour.57 Instead, we treat the ORDC curves as fixed with a maximum total price
adder of VOLL minus $500, which causes prices to rise to the cap of $9,000/MWh in shortages,
because $500 is the highest-price emergency generation resource we model. Higher-cost
demand-response resources will be triggered in response to high ORDC prices and therefore
prevent prices from going even higher, but do not affect the “marginal energy component” of
56
We assume that the CC reference unit is not capable of providing either spin or non-spin from an
offline position, although we assume that the CT reference unit is capable of providing non-spin from
an offline position.
57
See ERCOT’s implementation in ERCOT (2013c).
36 | brattle.com
price-setting. We model the ORDC curves out to a maximum quantity of 8,000 MW where the
prices are near zero, although they never drop all the way to zero.
These ORDC curves create an economic incentive for units to be available as spinning or nonspinning reserve, which influences suppliers’ unit commitment decisions. We therefore model
unit commitment in three steps: (1) a week-ahead optimal unit commitment over the fleet, with
the result determining which long-lead resources will be committed;58 (2) a four-hour ahead unit
commitment (updated hourly) with an updated fleet outage schedule, with the result
determining the preliminary commitment and decommitment schedules for combined cycle
units; and (3) an hourly economic dispatch that dispatches online baseload units, and can commit
10-minute and 30-minute quickstart units if energy and spin prices are high enough to make it
more profitable than remaining offline (similarly, if prices are not high enough these units will
economically self-decommit).59 Note that 10-minute quickstart units can earn spin payments
from an offline position while 30-minute quickstart units can earn non-spin payments from an
offline position. These resources will not self-commit unless doing so would result in greater
energy and spin payments (net of variable and commitment costs) than would be available from
an offline position. We use a similar logic to economically commit or de-commit units until the
incentives provided by the ORDC are economically consistent with the quantity of resources
turned on.
G. RESERVE MARGIN ACCOUNTING
Throughout this report we use reserve margin accounting conventions consistent with those in
ERCOT’s CDR report, as illustrated in the example calculation in Table 12. In particular, we
note that the peak load used to calculate the reserve margin is already reduced for PRD and LR
self-curtailments. Peak load is then explicitly reduced for TDSP energy efficiency and load
management, 10-minute and 30-minute ERS, and non-controllable LRs. On the supply side,
most resources are counted toward the reserve margin at their nameplate capacity, with wind
and solar counting at 8.7% and 100% of nameplate respectively, and the DC ties counting at 50%
of the path ratings.
58
Short-term resources are included in the week-ahead commitment algorithm, but their commitment
schedule is not saved since it will be dynamically calculated in a shorter window. But using short-lead
resources in the week-ahead commitment allows them to affect the commitment of long-lead
resources.
59
These week-ahead and day-ahead commitment algorithms minimize cost subject to meeting load as
well as ERCOT’s administratively-determined regulation up and spinning reserve targets, with nonspinning reserve targets not considered at the unit commitment phase.
37 | brattle.com
Table 12 Example Reserve Margin Calculation Reserve Margin Component
Quantity
(MW)
Grossed‐up Peak Load
Peak Load from ERCOT Shapes
TDSP Energy Efficiency Programs
71,159
70,618
541
Load Reductions
LRs serving RRS
10‐Minute ERS
30‐Minute ERS
TDSP Energy Efficiency Programs
TDSP Curtailment Programs
2,410
1,205
347
77
541
240
Supply
Conventional Generation
Hydro
Wind
Solar
Storage
PUNs
50% of DC Ties
76,658
69,700
521
1,319
124
36
4,331
628
Reserve Margin
11.5%
Sources and Notes: Reserve Margin = Supply/(Grossed‐up Peak Load – Load Reductions) – 1 Peak load implemented in modeling is different from the peak load used for calculated reserve margin because the modeled load shapes exclude TDSP EE gross‐ups, but include a separate gross‐up for 1.5% PRD and 195 MW of additional LRs not counted in the reserve margin. Conventional Generation includes seasonal, mothballed, and new units. Wind and solar contribute 8.7% and 100% of nameplate respectively. Although we use these CDR conventions to report reserve margins throughout this study, it is
worth noting, given the broader context of this study, how it might make sense to revise these
accounting conventions if the PUCT were to implement a mandatory reserve margin.
Whether through a centralized or bilateral capacity market, imposing a mandatory reserve
margin would make capacity valuable as a new product. This would substantially increase the
importance of reserve margin accounting, measurement, and verification measures. In that case,
it would be important to develop a well-defined capacity product where all types of supply
resources are interchangeable from a resource adequacy perspective, and include only resources
committed to supply capacity in ERCOT in the reserve margin calculation. Some of the revisions
that might be considered include:

Remove Uncommitted Supply. Currently, there are several types of non-firm
supply included in the reserve margin calculation, such as potential new builds
38 | brattle.com
and mothballed units. If a reserve margin mandate were implemented on a threeyear forward basis, it would require that all plants currently mothballed or under
development provide a firm commitment of being available in the delivery year.
Without such a commitment, they would not be included in the reserve margin
or earn capacity payments.

Revise Approach to Private Use Networks and Switchable Units. Similarly for
Private Use Networks (PUNs) and switchable units, these resources would have to
commit to providing resource adequacy in ERCOT before they could be
compensated or included at full value in the reserve margin calculations. For
PUNs, which tend to provide a combination of generation and demand response,
it may be necessary to separately and explicitly account for those networks’ gross
load, demand-response resources, and generation.

Revise Treatment of DC Ties. Interties that contribute to resource adequacy do
not constitute supply on their own. For that reason it may be beneficial to
consider adopting the approach used in other capacity markets of: (a) removing
DC Ties as a line item on the supply side that counts toward the reserve margin;
(b) enabling importers to contribute to the reserve margin and sell capacity into
ERCOT as long as they are supported by firm transmission rights; and (c) setting
aside a capacity benefit margin on the interties that would not be used for firm
imports but would instead be used to provide “tie benefits” or imports that are
probabilistically available to contribute to resource adequacy due to load diversity
with neighboring regions (and therefore reduce the required system reserve
margin).

Move from Installed to “Unforced” Plant Capacity Accounting. As is done in
MISO, PJM, and NYISO, consider moving from installed capacity (ICAP) to
unforced capacity (UCAP) accounting, which reduces each resource’s capacity
rating consistent with its own availability and outage rate. Similar to the effective
load carrying capability (ELCC) calculation done for wind, moving to UCAP
accounting would make the reserve margin contribution and economic payments
for each resource more consistent with its delivered reliability value. Some
special resource types such as demand response and storage might require their
own ELCC studies, to determine how substantially factors such as call limits affect
their reliability value as compared to generation.

Account for Committed Demand Response on the Supply Side. While demand
response can be accounted for on the supply or demand side, accounting in
centralized auctions is a bit more transparent if all types of capacity resources are
allowed to compete as supply-side resources. Interchangeable demand-response
accounting also requires considering the gross-up that is appropriately applied to
demand response to account for avoided line losses.
39 | brattle.com
Adopting these alternative accounting conventions would have no impact on reliability, but
would increase the transparency of reserve margin accounting and provide a more accurate
reflection of the expected resource adequacy value of each type of resource.
III. Reliability and System Costs Results
In this Section, we summarize our primary reliability and system cost results over a range of
assumed reserve margins. We report a variety of reliability metrics similar to those reported in
other reliability models, as well as total system cost results including capital, production, and
reliability event costs. We also report the sensitivity of these results to several modeling
assumptions including the assumed reference technology, forward period, likelihood of extreme
weather events, and marginal system costs.
A. SYSTEM RELIABILITY
As with other resource adequacy studies, including a recent study conducted for ERCOT, we
estimate realized reliability as a function of the planning reserve margin.60 We report here
several standard reliability metrics indicating firm load shed rates as a function of reserve margin,
the frequency of non-load-shed reliability events, and the sensitivity of these results to our study
assumptions.
1. Physical Reliability Metrics
Traditionally, ERCOT has determined its “target” reserve margin based on the 1-in-10 standard,
i.e., a probability-weighted average of 0.1 loss-of-load events (LOLE) per year. We report LOLE
as a function of reserve margin in Figure 18, with the dark blue line reflecting our Base Case
assumptions. The Base Case results show that a 14.1% reserve margin would be required to
achieve 0.1 LOLE. At that level, events would be expected to occur once per decade, each with
about 1,300 MW of load being shed for 2.3 hours on average.
Figure 18 also shows the sensitivity of this result to the probability of 2011 weather recurring
more frequently (left) and to the assumed multi-year forward period at which supply decisions
are locked in (right). Increasing the likelihood of 2011 weather from a 1% chance to a 1-in-15
year chance (with equal weighting on all other weather years) would increase the reserve margin
needed to meet the 1-in-10 standard from 14.1% to 16.1%.
The right-hand chart of Figure 18 shows the impact of assuming suppliers need less time to lock
in their supply decisions, which reduces the realized non-weather load forecasting uncertainty
and, therefore, the LOLE associated with a particular planning reserve margin. Reducing the
forward assumed lock-in period from three years in the Base Case assumption to one year would
60
See ECCO (2013a and b).
40 | brattle.com
reduce the reserve margin needed to meet the 1-in-10 standard from 14.1% to 13.1%. For
illustrative purposes, we also show that the 1-in-10 reserve margin would drop to 12.6% if it
were possible to completely eliminate non-weather load forecasting errors.
1.0
1.0
0.9
0.9
0.8
0.8
0.7
Base Case
1% Chance of
2011 Weather Equal 1/15 Chance of
Recurring
2011 Weather
Recurring
0.6
0.5
0.4
LOLE (Events/Yr) LOLE (Events/Yr) Figure 18 Loss of Load Events vs. Reserve Margin With Differing Weather Weights (Left) and Varying Forward Periods (Right) 0.3
4‐Year Forward Load Forecast Error
Base: 3‐Year 0.6
0.5
2‐Year 0.4
1‐Year 0.3
No LFE
0.2
0.2
0.1
0.7
1 Event in 10 Years
0.1
1 Event in 10 Years
0.0
0.0
9%
11%
13%
15%
Reserve Margin (ICAP %)
17%
9%
11%
13%
15%
Reserve Margin (ICAP %)
17%
Figure 19 summarizes resource adequacy as a function of reserve margin using three different
types of physical reliability metrics: (1) loss of load events (i.e., LOLE) on the left; (2) loss of load
hours (LOLH) in the middle; and (3) normalized expected unserved energy (normalized EUE) on
the right. As discussed above, the Base Case reserve margin required to yield a 0.1 LOLE is
14.1%. If, instead, a resource adequacy requirement of one “day” in 10 years is interpreted as 24
hours per 10 years, or 2.4 hours per year, the reserve margin would only need to be 9.1%. This
2.4 hours per year interpretation of the 1-in-10 standard is currently utilized by the Southwest
Power Pool.61
The far right panel on Figure 19 shows a third reliability metric, normalized EUE. Normalized
EUE is an alternative to the 1-in-10 standard that a NERC task force recently recommended to
address the limitations of traditional 1-in-10 LOLE and LOLH standards.62 It refers to the total
annual MWh of firm energy expected to be shed, divided by the total MWh of annual system
load. It represents the percentage of system load that cannot be served due to supply shortages.
61
SPP’s standard corresponds to a planning reserve margin of 13.6% (equivalent to what SPP refers to as
a 12%, capacity margin) although SPP’s current reserve margin currently is well above that level. See
SPP (2010).
62
See NERC (2010).
41 | brattle.com
Some international markets already use normalized EUE to set minimum reliability thresholds or
to trigger administrative interventions, although the metric may be referred to as Loss of Load
Probability (LOLP) or Unserved Energy (USE). Examples of metrics equivalent to normalized
EUE used in international markets include: (a) a 0.001% LOLP standard in Scandinavia; and (b) a
0.002% USE standard in Australia’s National Energy Market (NEM) and South West
Interconnected System (SWIS).63 As shown in Figure 19, using a normalized EUE of 0.001 as a
physical reliability standard in ERCOT would require a Base Case reserve margin of 9.6%.
We recommend adopting normalized EUE as a preferred reliability metric for setting the
reliability standard because it is a more robust and meaningful measure of reliability that can be
compared across systems of many sizes, load shapes, and other uncertainty factors. Such a crosssystem comparison is not meaningful for either LOLE or LOLH because neither metric considers
the MW size of the outage endured nor the size of the system itself. For example, a one-hour,
100 MWh outage event and a one-hour, 10,000 MWh outage event would be counted identically
under the LOLE and LOLH metrics, even though the load shed amount under the second outage
event is one hundred times greater in magnitude. Moreover, the 1-in-10 standard represents a
higher level of reliability in a large system than in a small system because neither the LOLE nor
the LOLH metric is normalized to system size.64 This means that a 100 MWh, one-hour outage
will have the same LOLE and LOLH values in a 10,000 MW and 100,000 MW power system,
even though individual customers in the smaller system are ten times more likely to endure an
outage. Normalized EUE avoids these shortcomings of the LOLE and LOLH metrics.
63
See Nordel (2009), p. 5; AEMC (2007), pp. 29-30, (2010), p. viii.
64
This is true as long as the one event represents a smaller proportion of total load in a large system than
in a small system.
42 | brattle.com
Figure 19 Reserve Margins Required to Meet Different Reliability‐Based Standards 1.0
14.1%
Reserve Margin
Requirement
0.9
8
7
9.1%
Reserve Margin
Requirement 0.003%
9.6%
Reserve Margin
Requirement LOLH (Hours/Yr)
LOLE (Events/Yr)
6
0.7
5
0.6
0.5
4
0.4
3
2.4 LOLH
0.3
2
0.2
0.1
0.1 LOLE
0.0
7%
Normalized EUE (% of MWh)
0.8
0.002%
0.001% Normalized EUE
0.001%
1
0
9% 11% 13% 15% 17% 7%
Reserve Margin (%)
0.000%
9% 11% 13% 15% 17%
Reserve Margin (%)
7%
9% 11% 13% 15%
Reserve Margin (%)
17%
Notes: Figure reflects reliability outcomes under our Base Case assumptions (3‐Year Forward LFE, 2011 Weather at 1% Probability).
Figure 20 illustrates the sensitivity of Base Case reliability results to the assumed probability of
individual weather years. The blue bars show the total MWh of annual load shed during each of
the 15 weather years for the Base Case simulations at a 14.1% reserve margin (consistent with
0.1 LOLE). As illustrated, the reoccurrence of 2011 weather would lead to almost 13,000 MWh
of expected involuntary curtailment of firm load, while there would be very little load being
shed in 12 out of the 15 weather years. The lines in Figure 20 show the probability-weighted
average of unserved energy assuming: (a) a 1% chance of 2011 weather reoccurring, as in our
Base Case; (b) a zero probability of 2011 weather; and (c) the probability of 2011 weather is equal
to those of the 15 other weather years. This chart highlights that assuming a higher probability
of sustained weather extremes substantially increases both the reliability-based and
economically-based reserve margin targets.
43 | brattle.com
Figure 20 Expected Unserved Energy by Weather Year at 14.1% Reserve Margin Expected Unserved Energy (MWh)
14,000
12,000
4,000
1/15 Chance of 2011 Weather
2,000
1% Chance of 2011 Weather
(Base Case)
0% Chance of 2011 Weather
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
1999
1998
0
Notes: Figure reflects Base Case 3‐Year forward LFE assumption, but bars are based on equal 1/15 weather weight for all years. Table 13 provides additional detail on how reliability varies with reserve margins. The left half
of the table shows LOLE, LOLH, and EUE across a range of reserve margins in our simulations.
As expected, these metrics show that annual outage rates decline with reserve margins. The
right half of the table reports the average outage duration, size, and depth for each load-shedding
event. It shows that the severity of events declines (along with the frequency) as reserve margins
increase.
44 | brattle.com
Table 13 Detailed Reliability Metrics across Planning Reserve Margins in Base Case Reserve
Margin
(%)
6.0%
7.9%
8.9%
9.8%
10.8%
11.8%
12.7%
13.7%
14.6%
15.6%
17.5%
Total Annual Loss of Load
LOLE
LOLH
EUE
(events/yr) (hours/yr)
(MWh)
2.51
1.36
0.91
0.64
0.44
0.29
0.18
0.12
0.08
0.04
0.01
7.99
4.02
2.62
1.77
1.19
0.74
0.46
0.28
0.18
0.08
0.03
16,402
7,555
4,750
3,020
1,921
1,145
664
370
229
97
29
Duration
(hours)
Average Outage Event
Energy Lost
(MWh)
Depth
(MW)
3.18
2.95
2.88
2.76
2.68
2.56
2.49
2.38
2.26
2.14
2.00
6,531
5,555
5,214
4,719
4,323
3,938
3,592
3,186
2,939
2,517
2,142
2,053
1,882
1,811
1,709
1,614
1,541
1,445
1,339
1,300
1,174
1,069
Notes: Table reflects reliability outcomes under our Base Case assumptions (3‐Year Forward LFE, 2011 Weather at 1% Probability). 2. Emergency Event Frequency
Figure 21 summarizes the frequency of six types of emergency events for the Base Case
simulations as a function of the installed reserve margin. The emergency events, in increasing
order of severity, are (1) the economic dispatch of emergency generation, (2) calling 30-minute
ERS, (3) calling TDSP load curtailments, (4) re-dispatching LRs from RRS to energy, (5) calling
10-minute ERS and, finally, (6) shedding firm load. As shown, at the 1-in-10 reserve margin of
14.1%, emergency generation will be dispatched approximately 5 times a year on a weightedaverage basis across all simulated years. At a reserve margin of 8.7%, the system faces one load
shed event per year on average, most years without load shed events and some years with
several. At the same 8.7% reserve margin, the various types of demand resources would have to
be called from three to eight times on average each year (depending on the resource type), and
emergency generation would be dispatched approximately 17 times on average each year.
All types of emergency events become more frequent at lower reserve margins, but the
frequency of re-dispatching LRs from RRS to energy increases faster than ERS and TDSP calls.
This is because at lower reserve margins the call-limited ERS and TDSP demand-side resources
call limits more often, meaning that their reliability value diminishes and ERCOT will need to
rely more heavily on other measures and resources.
45 | brattle.com
Figure 21 Average Annual Frequency of Emergency Events Event Frequency (events/yr)
30
0.1 LOLE
25
Emergency Generation
20
15
LRs
10
TDSP
30‐min ERS
10‐min ERS
Load Shed
5
0
7%
9%
11%
13%
Reserve Margin (% ICAP)
15%
17%
Notes: Results from Base Case (3‐Year Forward LFE, 2011 Weather at 1% Probability). B. ECONOMICALLY OPTIMAL RESERVE MARGIN
The primary result from our study is an estimate of the economically optimal reserve margin that
minimizes total system capital and production costs. We estimate that economic optimum to be
an approximate 10.2% reserve margin under our Base Case assumptions. We also discuss the
probability distribution of annual system costs across different levels of reserve margins.
1. System Cost-Minimizing Reserve Margin
Figure 22 summarizes the annual averages of total ERCOT reliability-related costs from our Base
Case simulations over a range of planning reserve margins. At each reserve margin level, we
show the weighted-average costs across all 7,500 annual simulations, with the individual cost
line items including:

Marginal CC Capital Costs are the annualized fixed costs associated with building
more CC plants, at a cost of $122.1/kW-yr in the Base Case, see Section II.C.5.

Production Costs (Above $10 billion per year Baseline) are total system
production costs of all resources above an arbitrary baseline cost of $10 billion.
We show only a portion of total system costs as an individual slice on the chart in
order to avoid having production costs dwarf the magnitude of other cost
components, and subtract the same $10 billion at all reserve margins shown.
Production costs decrease at higher reserve margins because adding efficient new
gas CCs reduces the need to dispatch higher-cost peakers.

External System Costs (Above Baseline) include production and scarcity costs in
neighboring regions above an arbitrary baseline, which drop by a small amount
46 | brattle.com
with increasing reserve margins because ERCOT will rely less on imports from
high-cost external peakers during internal scarcity events, and may also be able to
export more supply during external scarcity events.65

Emergency Generation is the price-driven dispatch of units outputting at high
levels above their summer peak ratings at an assumed cost of $500/MWh, see
Section II.F.3.

10-Minute and 30-Minute ERS is the cost of dispatching these resources during
emergency events at assumed costs of $3,681 and $1,405/MWh for 10-minute and
30-minute ERS respectively, see Section 0.

Non-Controllable LR costs reflect the cost of voluntarily self-curtailed LRs at a
cost of $380/MWh during high-price events, as well as the cost of
administratively re-dispatching LRs from supplying RRS to supplying energy at a
cost of $2,569/MWh during emergencies, see Section II.D.3.

TDSP Load Management costs are incurred when ERCOT administratively orders
these demand-side resources to curtail during emergencies at an assumed cost of
$2,450/MWh, see Section II.F.2.

Price Responsive Demand is the cost of voluntary self-curtailment among the 1%
of load resources that are assumed to respond to high-price events, which are
more common at lower reserve margins, see Section II.D.4.

Spinning and Non-Spinning Reserve Shortage costs are calculated as the area
under the ORDC curve, calculated assuming load would be shed at X = 1,150 MW,
see Section II.F.5.

Regulation Shortage costs are calculated according to the PBPC assuming that this
curve accurately reflects the marginal cost of running short on regulating reserves,
see Section II.F.4.

Firm Load Shedding costs are the customer costs imposed during load-shed events
at a cost at the assumed VOLL of $9,000/MWh.
At the lowest reserve margin shown in the chart, average annual reliability costs are high and are
driven by the high cost of emergency generation, external system costs, and other reliability
costs during scarcity conditions. As planning reserve margins increase, total reliability costs drop
more quickly than the increases in capital and production costs associated with adding additional
CCs. As a result, total costs drop as the reserve margin increases until the “economically optimal”
quantity of capacity has been added at a reserve margin of 10.2%. After crossing this minimum
cost point, the capital costs of adding more CCs exceed the benefits from reduced reliabilityrelated costs, and so total costs increase.
65
The baseline level of external production costs is not included in our total system cost. This differs
from our reporting of ERCOT-internal production costs, for which we do include baseline costs (that
do not vary with reserve margin) in order to produce a meaningful total cost estimate for the ERCOT
system.
47 | brattle.com
This 10.2% risk-neutral economically optimal reserve margin is substantially below the 14.1%
reserve margin based on the 0.1 LOLE standard but similar to the reserve margin that would be
required under the 2.4 LOLH and 0.001% normalized EUE standards.
Figure 22 Total System Costs across Planning Reserve Margins Total System Costs ($Million/Year)
$36,000
Firm Load Shedding
Regulation Shortages
Non‐Spinning Reserve Shortages
Spinning Reserve Shortages
Price‐Responsive Demand
TDSP Load Management
Non‐Controllable LRs
30‐Minute ERS
10‐Minute ERS
Emergency Generation
External System Costs (Above Baseline)
Production Costs (Above $10B/Yr Baseline)
Marginal CC Capital Costs
Economically Optimal Reserve Margin at 10.2%
$35,800
$35,600
$35,400
$35,200
$35,000
17.5%
16.5%
15.6%
14.6%
13.7%
12.7%
11.8%
10.8%
9.8%
8.9%
7.9%
7.0%
6.0%
$34,800
ERCOT Reserve Margin (% ICAP)
Notes: Total system costs include a large baseline of total system costs that do not change across reserve margins, including $15.2 B/year in transmission and distribution, $9.6 B/year in fixed costs for generators other than the marginal unit, and $10B/year in production costs. The total cost curve shown in Figure 22 has a shape similar to that which we have observed in
value-of-service studies for many other electric systems.66 The curve is relatively flat near the
minimum average cost point, indicating that expected total costs do not vary substantially
between reserve margins of 8% to 14%. However, as we discuss further below, the lower end of
that minimum cost range is associated with much more uncertainty in realized annual reliability
costs and a much larger number of severe, high-cost reliability events. At the 14% reserve
margin, a greater proportion of total annual costs is associated with the cost of adding CCs
(which has less uncertainty), and a smaller proportion of the average annual costs are from
uncertain, low-probability, but high-cost reliability events.
66
For example, see Poland (1988), p.21; Munasinghe (1988), pp. 5-7 and 12-13; and Carden,
Pfeifenberger, and Wintermantel (2011).
48 | brattle.com
2. Exposure to Extreme Shortage Events
The economic results shown in the previous sections assume risk neutrality with respect to the
uncertainty and volatility of reliability-related costs. This allows us to compare total costs at
different reserve margins in Figure 22 simply as the probability-weighted average of annual
reliability costs for all 7,500 simulation draws. However, there is substantial volatility around
the average level of possible reliability cost outcomes. Most simulated years will have very
modest reliability costs, while a small number of years have very high costs. These high-cost
outcomes account for the majority of the weighted-average annual costs shown as the individual
bars in Figure 22.
Figure 23 summarizes this risk exposure by comparing the weighted-average costs for different
reserve margins (shown as the individual bars in Figure 22) to annual costs under the most costly
possible outcomes, represented by the 75th, 90th, and 95th percentiles of annual reliability costs
across all 7,500 simulated scenarios. The figure shows that a substantial fraction of all reliabilityrelated costs are concentrated in the most expensive 10% of all simulation runs for each planning
reserve margin; for 10% of possible annual outcomes, the annual reliability costs are at or above
the 90th percentile line shown in the chart. While total average costs change by a relatively
modest amount over a range of planning reserve margins, differences in planning reserve
margins have a larger impact on the uncertainty in reliability costs and the likelihood of highcost outcomes than can be encountered in any particular year.
Considering the higher cost uncertainty exposure at lower reserve margins, some planners and
policymakers prefer to set planning reserve margins above the risk-neutral economic optimum.
As the simulation results show, a several percentage point increase in the reserve margin would
only slightly increase the average annual costs, but more significantly reduce the likelihood of
experiencing very high-cost events. As we will show in later sections, this mitigating impact is
modest when evaluated from a system cost perspective as we do here, but more substantial when
viewed from a customer cost or supplier net revenue perspective, as we discuss in Section IV.C
below.67
67
This is primarily because price * quantity changes much more substantially than does production cost
with small changes in quantity during scarcity events.
49 | brattle.com
Figure 23 Uncertainty Range in Total Annual System Costs Total System Costs ($Million/Year)
$40,000
Base Case Optimal RM
Base Case 0.1 LOLE $39,000
95th Percentile $38,000
90th Percentile $37,000
75th Percentile Average $36,000
Median $35,000
7%
9%
11%
13%
15%
17%
Reserve Margin (% ICAP)
Notes:
Total system costs include scarcity‐related and production costs (that decrease with reserve margin), generation capital costs (that increase with reserve margin), and T&D costs (which remain constant across reserve margins. Additional detail on the individual components of total system costs is available in Section III.B.1. For example, our Base Case simulations show that the risk-neutral, economically optimal
planning reserve margin is 10.2% compared to the 14.1% reserve margin needed to achieve the
0.1 LOLE standard. As Figure 22 and Figure 23 show, the increase in average annual system costs
required to achieve the 14.1% planning reserve margin (rather than the 10.2% economically
optimal reserve margin) is relatively modest at approximately $110 million per year.
Figure 23 also shows that this would reduce the annual system costs incurred once in a decade
(i.e., costs above the 90th percentile) by at least $196 million per year, and the costs incurred once
in 20 years (i.e., costs above the 95th percentile) by at least $313 million per year. In other words:
(a) a risk-neutral policymaker would not increase reserve margins above the 10.2% risk-neutral
optimum because, by definition, the expected costs would exceed expected benefits; (b) a
somewhat risk-averse policymaker might prefer slightly higher reserve margins but possibly not
high enough to meet 0.1 LOLE at a 14.1% reserve margin where the quantified incremental costs
exceed the quantified incremental benefits by a ratio of approximately 1.5-to-one; and (c) a more
risk-averse policymaker might wish to meet or even exceed the 14.1% reserve margin needed to
meet 0.1 LOLE. However, this discussion addresses only the risks of high system costs. Market
participants are more likely to care about risks to customer cost or supplier net revenues, which
we present in Section IV.C below.
50 | brattle.com
C. SENSITIVITY TO SYSTEM CONDITIONS AND STUDY ASSUMPTIONS
In this Section, we evaluate the sensitivity of economically optimal reserve margin estimates to a
variety of alternative study assumptions. We examine the impacts of: (a) changing the assumed
reference technology from a CC to a CT, (b) varying the assumed cost of building new plants,
(c) varying the forward period and associated load forecast uncertainty, (d) the assumed chances
of 2011 weather recurring, and (e) revising ORDC-based pricing such that prices are always
equal to marginal costs.
1. Marginal Resource Technology Type and Cost of New Entry
The Base Case simulations assume that a natural-gas-fired CC is the marginal resource that will
be added to increase reserve margins. In reality, it is more likely that a mix of gas CCs and gas
CTs may be added over the coming years. To evaluate the impact that the CC-based assumption
has on study results, Figure 24 compares the difference in total system costs for natural-gas-fired
CCs and CTs as the marginal resource. The chart shows very similar impacts on total system
reliability costs across the range of reserve margins for the two different resource types. The
greater capital costs of the CC are approximately balanced out by the greater production cost
savings, such that adding either resource type contributes approximately the same net value.
This indicates that the ERCOT system likely has a near-optimal mix of CCs and CTs in the
current fleet, which may also suggest that a mix of CCs and CTs would likely be built going
forward. However, the CT option is somewhat more economic overall in that it has a higher
economically optimal reserve margin, at 10.7% for the CT compared to 10.2% for the CC. The
total system cost implications are not very different for the two resource types, however, with
the minimum CT cost point being only about $32 million/year lower than the CC minimum cost
point.
51 | brattle.com
Figure 24 Total Annual System Costs with a Gas CC or CT Providing the Marginal Capacity $36,400
Total System Costs ($Million/Year)
$36,300
$36,200
$36,100
$36,000
Marginal CC (Base) $35,900
$35,800
Marginal CT
$35,700
$35,600
$35,500
7%
9%
11%
13%
Reserve Margin (% ICAP)
15%
17%
Notes:
Total system costs include scarcity‐related and production costs (that decrease with reserve margin), generation capital costs (that increase with reserve margin), and T&D costs (which remain constant across reserve margins. Additional details on the individual components of total system costs are available in Section III.B.1. We
also examine the impact of varying the gross CONE values of each type of marginal resource.
Figure 25 shows the impact of varying gross CONE from –10% to +25% relative to our base
assumption, with the CC on the left and CT on the right. Based on our experience, this range is
consistent with the uncertainty range behind technology cost estimates. Increasing the assumed
gross CONE value reduces the economically optimal reserve margin because the benefits of
achieving higher reserve margins no longer exceed the marginal costs. Overall, the economically
optimal reserve margin could vary over a range of 9.2%–11.5% depending on the assumed
marginal resource type and range of gross CONE uncertainty.
52 | brattle.com
$36,400
$36,400
$36,300
$36,300
High CC CONE
$36,200
$152.6/kW‐yr
$36,100
Base CC CONE
$36,000
$122.1/kW‐yr
$35,900
$35,800
Low CC CONE
$109.9/kW‐yr
$35,700
Cost‐Minimizing Reserve Margin
$35,600
Total System Costs ($Million/Year)
Total System Costs ($Million/Year)
Figure 25 Total Annual System Costs with Varying CC Gross CONE (left) and CT Gross CONE (right) $36,200
High CT CONE
$36,100
$121.3/kW‐yr
$36,000
Base CT CONE
$35,900
$97/kW‐yr
$35,800
$35,700
Low CT CONE
$87.3/kW‐yr
$35,600
$35,500
Cost‐Minimizing Reserve Margin
$35,500
7%
9%
11%
13%
Reserve Margin (% ICAP)
15%
17%
7%
9%
11%
13%
15%
17%
Reserve Margin (% ICAP)
Notes:
Total system costs include scarcity‐related and production costs (that decrease with reserve margin), generation capital costs (that increase with reserve margin), and T&D costs (which remain constant across reserve margins. Additional detail on the individual components of total system costs is available in Section III.B.1. 2. Forward Period and Load Forecast Uncertainty
As explained previously, non-weather load forecasting error (LFE) increases with the forward
period. This is unlike weather-related uncertainty, which is generally assumed to be constant
over time. A longer forward planning period therefore results in higher load forecast uncertainty
and, if resource additions cannot be modified on a shorter-term basis, a higher likelihood of
reliability and scarcity events. This means that a longer forward period will also increase the
reserve margin necessary to achieve any given reliability standard.
In our Base Case analysis, we assume that three years forward is the time at which all supply
decisions must be locked in, which is approximately consistent with the lead time needed to
construct new generation resources.68 This lead time for constructing new resources is also the
reason that PJM and ISO-NE’s capacity markets rely on a three-year forward period. In
traditionally-regulated regions with integrated resource planning processes subject to state
68
Note that although the entire development timeline for most new plants is greater than three years,
much of that development work, including siting and permitting, can be done without making major
irreversible financial commitments. This means that the time needed for actual plant construction is
most relevant when the resource investment decision is truly locked in. Developing and constructing
a natural gas CC or CT takes approximately 3.25 and 2.8 years respectively, see Spees, et al. (2011),
Appendices A.3 and B.3.
53 | brattle.com
regulation, at least a portion of all planning decisions are often made on a schedule that looks
more than three years forward.69 Note, however, that most systems also have substantial
flexibility to adjust their total resource portfolio on a shorter-term basis, including even a oneyear forward basis. For example, there is substantial capability to adjust DR commitments, adjust
retirement or retrofit decisions, delay or accelerate plant development efforts, and invest in plant
upgrades on a shorter-term basis.
We examine here the implications of varying the assumed forward period on the economically
optimal and reliability-based reserve margins. Generally, increasing the forward period at which
supply decisions are locked in increases the economic load forecast error, which consequently
increases the planning reserve margin whether the reserve margin is reliability-based or
economically-based.
Figure 26 shows that increasing the forward period will increase the economically optimal
planning reserve margin. The Base Case assumption of a three-year forward period is shown as a
blue line while the gray lines show costs ranging from a one-year to a four-year forward
planning period, as well as an illustrative case with no non-weather LFE. Total costs increase
with a higher forward period because there is a greater risk of low-reliability events associated
with under-forecasting load (and greater capital costs required to build sufficient capacity to
avoid such events). However, if sufficient short-term resource flexibility exists to allow for a
reduction of the forward planning period from three years to zero, then the risk-neutral optimal
reserve margin would decrease by 1.1%, from 10.2% to 9.1%.
69
For example, see the discussion of the Long-Term Resource Planning processes implemented by
California’s investor owned utilities under the oversight of the state commission in Pfeifenberger, et
al. (2012).
54 | brattle.com
Figure 26 Total Annual System Costs with different Forward Periods for Locking in Supply Decisions $36,400
Total System Costs ($Million/Year)
$36,300
$36,200
$36,100
4‐Year Forward Load Forecast Error
$36,000
$35,900
Base: 3‐Year $35,800
$35,700
2‐Year
1‐Year $35,600
No LFE
Cost‐Minimizing Reserve Margin
$35,500
7%
9%
11%
13%
Reserve Margin (% ICAP)
15%
17%
Notes:
Total system costs include scarcity‐related and production costs (that decrease with reserve margin), generation capital costs (that increase with reserve margin), and T&D costs (which remain constant across reserve margins. Additional detail on the individual components of total system costs is available in Section III.B.1. The same relationship between forward period and reserve margins also holds true for reliabilitybased reserve margins as summarized in Table 14 for 0.1 LOLE, 2.4 LOLH, and 0.001%
normalized EUE. For example, reducing the forward planning period from three years to one
year reduces the 0.1 LOLE-based reserve margin by 1.0 percentage point from 14.1% to 13.1%.
This would, of course, require that sufficient short-term resources exist that could be mobilized
on a 12-month basis should economic growth prove greater than anticipated.70 Such short-term
resources might include additional demand response (assuming the market is not fully saturated),
upgrades to existing plants (or new plants under construction), reactivations of mothballed
plants, an increase in net import commitments, imports (if they are qualified to provide capacity)
or the acceleration of in-service dates of new plants under development.
70
For example, assume that most resource investment decisions (e.g., for existing generation) were made
on a longer term basis but the final 5% of resources were not procured until one year prior to delivery.
However, the actual amount of incremental resources needed on a one-year forward basis may only be
3% or as high as 7% if economic growth were higher or lower than expected in prior planning
exercises.
55 | brattle.com
Table 14 Physical Reliability‐based Reserve Margins for Varying Forward Periods Load Forecast Error
No LFE
1 Year
2 Years
3 Years (Base)
4 Years
Reliability‐Based Reserve Margin
0.1 LOLE
2.4
LOLH
0.001% Norm. EUE
12.6%
13.1%
13.6%
14.1%
14.6%
8.2%
8.4%
8.7%
9.1%
9.5%
8.4%
8.7%
9.1%
9.6%
9.9%
3. Probability Weighting of Weather Years
Figure 27 shows the sensitivity of our economically optimal reserve margin estimate to the
likelihood of 2011 weather reoccurring. As the figure shows, the Base Case assumption of a 1%
probability results in a 10.2% economically optimal reserve margin. At a zero probability, the
optimal reserve margin is somewhat lower at 9.7%. However, if 2011 had a probability equal to
that of the other 14 weather years (i.e., a 1/15 chance), the economically optimal reserve margin
would increase to 11.5%. Thus, the economically optimal reserve margin estimate may change
by more than two percentage points depending on one’s view of the likelihood of such extreme
weather recurring.
56 | brattle.com
Figure 27 Total Annual System Costs with Varying Probability of 2011 Weather Recurring $36,400
Total System Costs ($Million/Year)
$36,300
Equal Chance of
2011 Weather
$36,200
$36,100
$36,000
Base: 1% Chance of 2011 Weather
$35,900
$35,800
0% Chance of 2011 Weather
$35,700
Cost‐Minimizing Reserve Margin
$35,600
$35,500
7%
9%
11%
13%
15%
17%
Reserve Margin (% ICAP)
Notes:
Total system costs include scarcity‐related and production costs (that decrease with reserve margin), generation capital costs (that increase with reserve margin), and T&D costs (which remain constant across reserve margins. Additional detail on the individual components of total system costs is available in Section III.B.1. 4. Energy Prices Always Equal to Marginal Cost
As explained in Section II.F above, our Base Case analysis simulates a complex scarcity pricing
structure; the most important component of which is the administrative pricing in the ORDC
curve. We implement an ORDC pricing function calculated as if load would be shed (or other
emergency actions undertaken with an equivalent cost equal to the value of lost load) at an
operating reserve level of X = 2,000 MW, consistent with ERCOT’s proposed implementation.
This creates scarcity prices that exceed marginal costs in some cases, because we assume that
ERCOT will not actually shed load or otherwise incur such high costs until operating reserves are
depleted to 1,150 MW. Emergency actions other than load shedding do incur costs between
X=2,000 and X=1,150, but we model those actions explicitly at costs less than VOLL, as described
in Section II.F.2.
To evaluate the impact of these prices in excess of marginal cost, we examine an alternative
Perfect Energy Price Case, where we assume that prices are always set equal to marginal cost
with the ORDC set at X = 1,150 MW for both pricing and marginal cost purposes. Another
minor difference between the two cases is that we do not reduce the maximum PBPC to the
57 | brattle.com
LCAP after PNM is exceeded, but instead assume that the PBPC remains at the same price that is
reflective of the marginal cost of enduring regulating reserve shortages.
Figure 28 shows the difference in realized system costs in the two cases across the range of
reserve margins. The chart shows that system costs would be lower under the Perfect Energy
Price Case, consistent with the expectation that having perfectly efficient energy prices will
result in the most efficient system dispatch and lowest overall system costs. This also reduces the
economically optimal reserve margin by almost a percentage point, from 10.2% to 9.3%. The
total system cost is $77 million/year lower on average at the optimum in the Perfect Energy Price
Case compared to the Base Case.
Figure 28 Total Annual System Costs in Base and Perfect Energy Price Cases (X = 2,000 MW in Base Case and X = 1,150 MW in Perfect Energy Price Case) $36,400
Total System Costs ($Million/Year)
$36,300
$36,200
$36,100
$36,000
Base Case
ORDC w/ X=2,000
$35,900
$35,800
Perfect Energy Price
ORDC w/ X=1,150
$35,700
$35,600
Cost‐Minimizing Reserve Margin
$35,500
7%
9%
11%
13%
Reserve Margin (% ICAP)
15%
17%
Notes:
Total system costs include scarcity‐related and production costs (that decrease with reserve margin), generation capital costs (that increase with reserve margin), and T&D costs (which remain constant across reserve margins. Additional detail on the individual components of total system costs is available in Section III.B.1. Table
15 summarizes the components of total system costs that contribute to the $77
million/year cost difference between the two cases. The most important factors driving the
higher cost in the Base Case relative to the Perfect Energy Price Case is the increased frequency
of PRD self-curtailments caused by higher prices and the increase in marginal CC capital costs
incurred to achieve a higher reserve margin and avoid some of those PRD costs. Other minor
contributing factors are greater imports and increased frequency of PBPC-based regulation
shortages.
58 | brattle.com
Some components of total system costs are actually lower in the Base Case, although not enough
lower to reduce system costs overall. Those reduced costs include: (a) lower production costs, as
some internal resources ramp down to provide reserves or are displaced by PRD and imports;
(b) PRD calls displacing the need to make administrative calls on LRs, TDSP load curtailments,
and ERS; (c) a reduction in reserve shortages; and (d) avoided load shedding, which is made
possible because avoiding some administrative DR calls, as previously mentioned, reduces the
likelihood of hitting their call limits and therefore necessitating load shed.
Table 15 Comparison of Total Annual System Costs in the Base and Perfect Energy Price Cases (Base Case at 10.2% and Perfect Energy Price Case at 9.3% Reserve Margin) Reliability‐Related Cost Component
Base Case at Economic Optimum
($M/Yr)
Marginal CC Capital Costs
Production Costs (Above Baseline)
External System Costs (Above Baseline)
Perfect Energy Cost Increase:
Base Case ‐ Price Case
at Optimum Perfect Energy Price Case
($M/Yr)
($M/Yr)
$510
$175
$100
$2
$3
$1
$15
$2
$57
$29
$34
$8
$27
$434
$196
$91
$2
$5
$1
$27
$3
$13
$31
$36
$4
$42
Transmission and Distribution
Generation Fleet Fixed Costs Production Costs Baseline
$15,160
$9,593
$10,000
$15,160
$9,593
$10,000
Total System Costs $35,716
$35,638
Emergency Generation
10‐Minute ERS
30‐Minute ERS
Non‐Controllable LRs
TDSP Load Management
Price‐Responsive Demand
Spinning Reserve Shortages
Non‐Spinning Reserve Shortages
Regulation Shortages
Firm Load Shedding
$76
($21)
$9
$0
($2)
($0)
($11)
($1)
$44
($2)
($2)
$4
($15)
$0
$0
$0
$77
Notes: Additional detail on the individual components of total system costs is available in Section III.B.1. D. SENSITIVITY OF ECONOMIC RESERVE MARGIN TO STUDY ASSUMPTIONS
Our estimate of the risk-neutral economically optimal reserve margin is sensitive to a number of
study assumptions as we have explained in the previous sections, and summarized in Table 16
below. As shown in the table, the economically optimal reserve margin for the most part ranges
from approximately 9% to approximately 12%, depending on study assumptions that drive
economic costs and the frequency of scarcity events. In addition to the drivers of scarcity that
59 | brattle.com
we have already discussed in prior sections, we also provide illustrative calculations in which we
vary VOLL and DR costs. We estimate that halving or doubling all VOLL-related and DRrelated costs would affect the economically optimal reserve margin by about +/-1.5%.
Table 16 Sensitivity of the Economically Optimal Reserve Margin to Study Assumptions Reserve Margin Range
(%ICAP)
Base Case
Marginal CC w/ X=2,000 MW
Vary Load Forecast Error
Vary CC CONE
Vary 2011 Weather Weight
Vary VOLL and DR Costs
Perfect Energy Price
Marginal CC w/ X=1,150 MW
Vary Load Forecast Error
Vary CC CONE
Vary 2011 Weather Weight
Vary VOLL and DR Costs
CT as Marginal Technology
Marginal CT w/ X=2,000 MW
Vary Load Forecast Error
Vary CC CONE
Vary 2011 Weather Weight
Vary VOLL and DR Costs
Base Assumptions
Low/High Sensitivity
3 Yrs
$122.1/kW‐yr
1%
See Section II.D.1
0 Yrs ‐ 4 Yrs
$109.9 ‐ $152.6/kW‐yr
0 ‐ 1/15
50% ‐ 200% of Base Cost
3 Yrs
$122.1/kW‐yr
1%
See Section II.D.1
0 Yrs ‐ 4 Yrs
$109.9 ‐ $152.6/kW‐yr
0 ‐ 1/15
50% ‐ 200% of Base Cost
3 Yrs
$97/kW‐yr
1%
See Section II.D.1
0 Yrs ‐ 4 Yrs
$87.3 ‐ $121.3/kW‐yr
0 ‐ 1/15
50% ‐ 200% of Base Cost
10.2%
9.1% ‐ 10.5%
9.2% ‐ 10.6%
9.7 ‐ 11.6%
8.9% ‐ 11.8%
9.3%
8.9% ‐ 9.5%
7.9% ‐ 9.8%
9.2% ‐ 10.6%
7.2% ‐ 10.8%
10.7%
10% ‐ 10.8%
9.8% ‐ 11.5%
10.3% ‐ 11.9%
9.0% ‐ 12.6%
Notes: VOLL and DR Sensitivities are approximate calculations not corresponding to simulation results.
IV. Comparison of Energy-Only and Capacity Market Designs
One pressing question before the Commission is whether ERCOT should maintain its current
energy-only market or implement a capacity market. The essential difference is that in an
energy-only market, the reserve margin and resulting reliability implications are determined by
market forces. Although the Commission can estimate the likely future reserve margin and
influence it by adjusting scarcity pricing provisions, the market will ultimately move toward an
“equilibrium” reserve margin at which prices are just high enough to attract investments in new
resources. By comparison, in a capacity market, the Commission would mandate the reserve
margin consistent with its reliability, economic, and other policy objectives, and market forces
would then determine the “equilibrium” capacity price that can sustain the investments needed
to achieve the mandated reserve margin.
60 | brattle.com
In this Section, we utilize our simulation modeling results to inform the policy question of
whether to maintain the current energy-only market or move to a mandated reserve margin and
capacity market design. For each of the two market designs, we evaluate the: (a) equilibrium
market prices (as opposed to short-term transitional conditions that might be experienced if
implemented or continued today); (b) reserve margins each design is likely to achieve; (c) yearto-year variations around equilibrium points, as well as uncertainty in the equilibria themselves;
and (d) consequences for customer costs and supplier net revenues on average over many years
and in extreme years. Finally, we evaluate the implications of these results for the policy
questions facing the Commission in determining the best course of market design into the future.
A. ENERGY-ONLY MARKET RESULTS
We describe here the equilibrium conditions that we would anticipate under ERCOT’s current
energy-only market design by: (1) estimating the energy-only market equilibrium for our Base
Case assumptions and several sensitivity cases; (2) summarizing the volatility in realized prices
and net revenues across reserve margins; and (3) describing the likely year-to-year variation in
realized reserve margins.
1. Equilibrium Reserve Margin
In an energy-only market, there is no mandatory reserve margin or associated reliability level.
Instead, the level of generation investment and corresponding reserve margin depends on market
prices. Investors build generation whenever expected future energy prices are high enough to
provide an adequate return on capital. If the reserve margin is very low, frequent shortage
conditions will lead to high expected prices, increased investment in new generation, and higher
reserve margins. However, if the reserve margin becomes too high, prices will not be sufficient
to support investment. Thus, the market will reach an equilibrium reserve margin where
suppliers are recovering their investments and earn an adequate return, but no more. We
illustrate such an equilibrium point in Figure 29, which we estimate at 11.5% in our Base Case.
At that point, the net revenues for a new combined-cycle plant (shown in red) are just equal to
its annualized capital and fixed costs at CONE (shown in blue).
61 | brattle.com
Figure 29 Average Annual Energy Margins for a Combined‐Cycle Plant $350
Base Case
0.1 LOLE
Base Case
Optimal RM
CC Energy Margins ($/kW‐yr)
$300
$250
Base Case
$200
$150
CC CONE
$100
Energy‐only equilibrium RM is higher than optimum
because prices sometimes exceed marginal cost
$50
$0
7%
9%
11%
13%
15%
17%
Reserve Margin (ICAP %)
Notes: Figure reflects base case assumptions, resulting in an energy‐only equilibrium reserve margin of 11.5%. We provide a comparison of this energy-only reserve margin and the associated reliability results
for the Base Case and various sensitivity cases in Table 17. Our Base Case market equilibrium
estimate of 11.5% is above the 10.2% economically optimal reserve margin and below the 14.1%
1-in-10 reserve margin we estimated in Section III above. This 11.5% market equilibrium
exceeds the 10.2% economically optimal reserve margin because the Base Case ORDC produces
energy prices that sometimes exceed marginal system cost (as explained in Section II.F) and,
therefore, provides investment incentives that slightly exceed the resource’s true economic
value. In the alternative Perfect Energy Price Case, where prices are always equal to marginal
cost, there is no such discrepancy, and the energy-only market reaches equilibrium exactly at the
lower system-wide cost-minimizing reserve margin of 9.3%.
If investors have different beliefs about probability distributions around load and other factors
affecting revenues or costs, the market equilibrium will differ from our estimates, as illustrated in
Table 17. Changing our assumptions about the likelihood of 2011 weather recurring, the level of
load forecast error, the marginal resource technology, and the ORDC scarcity pricing function
results in equilibrium reserve margins ranging from 9% to 13%. The actual uncertainty could be
even wider, however, when considering other variables such as natural gas price uncertainty, the
cost of new entry (CONE), different beliefs about potential regulatory interventions, etc.
This range of equilibrium reserve margins would produce a range of reliability outcomes, which
we estimate to be 0.27–0.85 LOLE, 0.68–2.37 LOLH, and 0.0003%–0.0013% normalized EUE.
Thus, the energy-only market will result in an equilibrium reliability level that is below the
62 | brattle.com
traditional 1-in-10 LOLE standard, but at a level consistent with possible alternative reliability
standards, including the 2.4 LOLH standard used in SPP and the 0.001% normalized EUE
standard used in some international markets.
Table 17 Comparison of Energy‐Only Equilibrium to Alternative Reserve Margin Targets Scenario
Reliability at Energy‐Only Equilibrium
Economic Energy‐Only 0.1 LOLE
Optimum Equilibrium
Base Case
Equal Chance of 2011 Weather
No Non‐Weather Forecast Error
Perfect Energy Price
CT as the Marginal Technology
10.2%
11.5%
9.4%
9.3%
10.7%
11.5%
12.9%
10.8%
9.3%
11.6%
LOLE
(Events/Yr)
LOLH
(Hours/Yr)
Norm. EUE
(%)
0.33
0.43
0.27
0.85
0.30
0.86
1.15
0.68
2.37
0.79
0.0004%
0.0005%
0.0003%
0.0013%
0.0003%
14.1%
16.1%
12.6%
14.3%
14.1%
Notes: Reliability is slightly better in the Base Case than in the Perfect Energy Price Case because the higher prices created by ORDC attracts more price responsive demand and reduces the need to rely on call‐limited administrative demand‐response resources, see Section III.C.4. 2. Volatility in Realized Prices and Energy Margins
Our estimate of the energy-only equilibrium reserve margin is strongly influenced by our
assumed probability distributions for peak loads and generator outages, especially the most
extreme scarcity events at the tails of those distributions. As the reserve margin declines, these
tails become more likely to produce shortages, high prices, high system-wide costs, and high
generator margins. We analyzed these effects by simulating the entire probability distributions
for each possible reserve margin.
Figure 30 shows the resulting range of annual values for the Base Case. The upper percentile
curves show that annual prices and supplier net revenues in the tails of the distribution can be
much higher than the median year or the overall weighted average. This indicates that the highcost tails have a substantial effect on the energy-only market equilibrium reserve margin.
For example, at the Base Case equilibrium reserve margin of 11.5%, we estimate that energy
prices would exceed $74.3/MWh (50% higher than the typical or “median” price) once per
decade (90th percentile) and would exceed $88.7/MWh (70% above the median price) once every
two decades (95th percentile). Similarly, although supplier net revenues are at CONE on average
across all years, the typical or median year has net revenues of only $97.2/kW-year (only 80% of
CONE), with net revenues exceeding CONE only once every 2.5 years. Suppliers will not be
earning sufficient net revenues to recover their investment costs most of the time and will
depend heavily on the net revenues realized during shortage years that would likely occur only a
few times over the asset life. Assuming full exposure to spot market prices (i.e., no hedging) net
revenues of generating plants would exceed $196/kW-year (1.6 times CONE) once in a decade
(90th percentile) and $263/kW-year (2.2 times CONE) once every two decades (95th percentile).
63 | brattle.com
Figure 30 Uncertainty Range in Realized Load‐Weighted Energy Prices (Left) and CC Energy Margins (Right) $200
$180
$160
$120
$100
$80
Average $60
$40
5th Percentile
95th Percentile
90th Percentile
$400
95th Percentile
90th Percentile
75th Percentile
$140
Energy‐Only Equilibrium
$450
CC Energy Margin ($/kW‐yr)
Energy Price ($/MWh)
$500
Energy‐Only Equilibrium
Median $350
75th Percentile
$300
$250
$200
Average $150
CC CONE
$100
$50
$20
5th Percentile
Median $0
$0
7%
8%
9% 11% 12% 13% 14% 15% 17%
7%
Reserve Margin (% ICAP)
8%
9%
11% 12% 13% 14% 15% 17%
Reserve Margin (% ICAP)
Some opponents of energy-only markets have asserted that such high spot market payoffs occur
too infrequently for investors to rely on. However, this argument ignores the fact that most
generators will sell their output at (e.g., seasonally hedged) forward prices, not spot prices. For
example, even seasonal forward prices will largely eliminate the weather-driven component of
spot price uncertainty. However, even if hedged through forward contracts, suppliers will still
face significant uncertainty from non-weather factors, as discussed in Section IV.C.2 below.
3. Year-to-Year Reserve Margin Variability
One of the main sources of uncertainty is load growth. Our Base Case simulations assume that
the market invests to exactly meet the equilibrium planning reserve margin on a three-year
forward basis. However, actual load growth will always differ from three-year expectations,
resulting in a range of realized reserve margins that differ from equilibrium reserve margins. We
simulate this effect based on the assumed probability distributions for non-weather forecast error
as described in Section II.B.3 above. This yields the range of planning and realized reserve
margins summarized in Figure 31. The three right-hand bars in the figure show planning reserve
margins estimated on a forward basis against the weather-normalized (50/50) peak load, while
the bar on the left shows the uncertainty range of realized reserve margins as measured after the
fact, considering both realized weather and non-weather uncertainties.
The chart shows that even if the three-year-ahead planning reserve margin is exactly at the
economic equilibrium of 11.5%, realized shorter-term planning reserve margins can be higher or
lower as load growth uncertainty resolves itself over the next three years. As the bar second
from the left shows, planning reserve margins projected going into each summer would thus vary
64 | brattle.com
around the equilibrium from 9.9%–13.2% in 50% of all years and drop below 8.5%
approximately once per decade (i.e., below the 10th percentile shown). Once weather-related
load fluctuations are considered as well, after-the-fact realized reserve margins will vary even
more substantially. As the bar on the very left shows, such after-the-fact realized reserve
margins will drop below 5.2% approximately once per decade (i.e., below the shown 10th
percentile). However, realized reserve margins, particularly the lows that largely reflect realized
weather extremes, should not be compared to more familiar planning reserve margin
benchmarks.
Figure 31 Year‐to‐Year Variability in Planning and Realized Reserve Margins 20%
18%
Reserve Margin (%)
16%
Realized Ratio of Installed Capacity to Peak Load
Based on Realized Weather
Planning Reserve Margins
Estimated on a Forward Basis Before Weather is Known
90th Percentile
14%
50% of All Years
12%
Market Equilibrium
3‐Year Ahead Planning Reserve Margin when Supply Decisions are Locked in
10%
8%
6%
10th Percentile
4%
Realized Reserve Margin
0 Yrs Ahead
1 Yr Ahead
2 Yrs Ahead
Planning Reserve Margin
3 Yrs Ahead
Actual variability in reserve margins may differ from the three-year-forward simulation results
shown in Figure 31. Our simulations do not account for the mitigating effect of short lead-time
resources (such as uprates and demand response) that can exit or enter the market as expectation
change between three years forward and delivery. By not simulating the effects of market exit
and entry by short-term resources, our results would tend to overstate the range of realized
reserve margins. However, our simulations also do not account for the countervailing effects of
additional supply-side uncertainties, such as unanticipated retirements, construction delays, and
lumpiness in uncoordinated new entry, which would tend to increase the variability of reserve
margins; and uncertainties about modeling assumptions, anticipated fuel prices, and other factors
that would further widen the distribution of realized reserve margins. Overall, we estimate that
equilibrium reserve margins would range from 9% to 13%, with an additional year-to-year
planning reserve margin variation of approximately 3 percentage points around that equilibrium.
65 | brattle.com
B. CAPACITY MARKET RESULTS
If ERCOT or the Commission wanted to achieve a higher or more certain reserve margin than
the energy-only market would support, then a mandatory reserve margin could be imposed at a
level consistent with such policy objectives. For example, a 14.1% reserve margin mandate
would correspond to the traditional 1-in-10 target for our Base Case simulation results.
Alternatively, the Commission could opt to specify a different LOLE (or EUE-based) target, or
even opt to define a new type of reserve margin requirement based on economic cost and risk
mitigation objectives.
1. Capacity Prices at Higher Reserve Margins
Mandating a higher reserve margin would make capacity valuable beyond the value obtained in
the energy market. This would make additional payments available to suppliers, either through a
centralized or bilateral capacity market.71 The capacity price consistent with market conditions
under any particular reserve margin mandate would be determined by market forces. In
equilibrium, the average capacity price would equal the net cost of new entry (Net CONE), equal
to the difference between CONE and the energy margins obtainable at the required reserve
margin.
Note that these estimates describe long-term equilibrium prices. Near-term prices might be
lower if the market has excess capacity or if low-cost resources (such as demand response and
generation uprates) are sufficient to meet the requirement at a lower price.
Figure 32 shows how expected equilibrium capacity prices would vary with reserve margins
under our Base Case assumptions. Based on our Base Case simulation results, setting a reserve
margin mandate below 11.5% would produce capacity prices of zero, since suppliers would earn
energy margins in excess of CONE and need no additional revenues to invest.72 However,
reserve margin mandates above the energy-only market equilibrium of 11.5% would lead to
positive capacity prices. The equilibrium average capacity price would increase as reserve
margins increase, since declining average energy prices would increase Net CONE. A mandate
equal to the traditional 1-in-10 LOLE standard at a 14.1% reserve margin would yield an average
capacity price of approximately $40/kW-year at equilibrium. These increasing reserve margins
71
We assume for the purposes of this report that the payments would be awarded through a capacity
market, but the level of capacity payments required would be the same under a range of alternative
capacity payment mechanisms (although some approaches might award these over a different time
schedule), as long as the approach does not involve price discrimination. For a comprehensive
description of resource adequacy standards, capacity markets, and alternative market designs, see
Pfeifenberger, et al. (2009).
72
In reality, prices would unlikely be literally zero, but instead may be at a very low level consistent
with any incremental overhead costs or penalty risks imposed by taking on a capacity obligation.
66 | brattle.com
would also correspond to energy prices and energy margins lower than in the energy-only
equilibrium, such that suppliers’ all-in net revenues equal CONE in both cases.
Note that these estimates describe long-term equilibrium prices. Near-term prices might be
lower if the market has excess capacity or if low-cost resources (such as demand response and
generation uprates) are sufficient to meet the requirement at a lower price.
Figure 32 Equilibrium Capacity Prices Required to Sustain Higher Reserve Margins Equilibrium Capacity Price ($/kW‐y)
$140
Base Case Optimal RM
0.1
LOLE
$120
CC CONE
$100
CC Energy
Margins
$80
$60
$40
Equilibrium Capacity Price $20
Net CONE = CONE ‐ Energy Margins
$0
7%
9%
11%
13%
15%
17%
Reserve Margin (% ICAP)
Figure 33 below shows equilibrium capacity prices for several alternative cases reflecting
different study assumptions. Cases with higher energy prices and energy margins yield lower
capacity prices, consistent with long-run total net revenues equal to CONE on average. At the
14.1% reserve margin needed to meet the 1-in-10 standard in the Base Case (but not in all
sensitivity cases), equilibrium capacity prices would vary from approximately $20 to $60/kWyear depending on energy market design and varying with study assumptions.
67 | brattle.com
Figure 33 Equilibrium Capacity Prices Under Varying Study Assumptions Equilibrium Capacity Price ($/kW‐y)
$140
Base Case Optimal RM
Base Case 0.1 LOLE
$120
CC CONE
$100
$80
$60
Perfect Energy Price
$40
No LFE
Base Case
$20
Equal Chance of
2011 Weather
$0
7%
9%
11%
13%
Reserve Margin (% ICAP)
15%
17%
2. Capacity Price Volatility
The above figures show only the equilibrium capacity prices that one would expect on average.
Actual capacity prices would vary from year to year around these levels. Energy price
uncertainty would be less in a market design with a higher mandated reserve margin than in an
energy-only market. This is because the higher reserve margin reduces energy market volatility
and average energy prices as discussed above. Although only expected future energy prices are
likely to be incorporated into capacity-market supply offers (through their impact on Net
CONE), capacity prices would nevertheless vary based on the supply and demand for capacity
during a particular delivery year. In fact, capacity prices can be quite sensitive to even small
shifts in supply and demand balances, because both supply and demand curves tend to be quite
steep. This is especially true on the supply-side in capacity markets without multi-year forward
procurement, which would have a near-vertical supply curve since all supply decisions have
already been made and suppliers have insufficient time to respond to changes in demand.73
We provide indicative estimates of such variations based on Monte Carlo simulations that we
conducted for another region’s capacity market, consistent with realistic shocks to supply and
demand, empirically-based three-year forward supply elasticities, and an assumed price cap of
73
For a more comprehensive discussion of volatility in capacity market prices, see Spees, et al. (2013).
68 | brattle.com
two times Net CONE (which limits volatility).74 These indicative simulation results are shown in
Figure 34. If ERCOT were to adopt a required reserve margin and a three-year forward capacity
market, pricing patterns may be different, depending on its own market characteristics and
market conditions.
Two of the most important factors affecting the level of volatility in a capacity market are: (1) the
forward period, with longer-forward periods reducing capacity price volatility by increasing
supply elasticity, and (2) the steepness of the demand curve, with a gradually sloped curve
reducing capacity price volatility relative to a vertical demand curve that reflects a fixed
requirement. However, it is important to note that some (but not all) factors reducing price
volatility can increase quantity volatility.
Figure 34 Approximate Uncertainty Range of Three‐Year Forward Capacity Prices Equilibrium Capacity Price ($/kW‐y)
$140
CC CONE
$120
95th Percentile
90th Percentile
75th Percentile
$100
$80
$60
Average $40
Median $20
5th Percentile
$0
7%
8%
9%
11%
12%
13%
14%
Reserve Margin (% ICAP)
15%
17%
C. IMPLICATIONS FOR CUSTOMERS AND SUPPLIERS
In this Section, we compare alternative energy-only and capacity market designs from the
perspectives of customers and suppliers. On the customer side, we evaluate how energy costs
and capacity costs, (if applicable) differ between the two market designs. On the supply side, we
evaluate the proportion of net revenues made up from energy margins versus capacity payments,
74
See the simulated distribution of capacity prices around Net CONE with the “Initial Candidate
Demand Curve”, from Newell, et al. (2014), p. 14.
69 | brattle.com
although the total payments at equilibrium are equal to CONE in both cases. Finally, we
evaluate the volatility profile of costs and revenues under either market design by looking at the
once-per-decade scarcity year, and reporting the realized total costs or net revenues in that
extreme event if the market participant is totally exposed to spot prices or if they are 80% hedged
against extreme weather.
1. Total Customer Costs and Volatility
In ERCOT’s competitive retail environment, the generation component of retail rates will reflect
wholesale market prices, which depend on market conditions and regulatory drivers such as gas
prices, realized reserve margins, and mandated planning reserve margins (if any). Very low
planning reserve margins would lead to frequent price spikes and high electricity rates; very high
reserve margins would depress energy prices but could only be sustained with high capacity
payments in the long run. Note that the current market is not in long-run equilibrium, with
reserve margins above our estimates of economically-sustainable reserve margins despite the
absence of capacity payments. Suppliers are currently earning less than CONE, and customers
are enjoying rates below a long-run sustainable level. By contrast, our equilibrium analysis
assumes that reserve margins above the energy-only equilibrium would require capacity
payments equal to Net CONE so suppliers would earn a total of CONE from the combination of
capacity payments and energy margins.
To determine the reserve margin that would minimize customer costs, we calculated equilibrium
energy and capacity prices (if any) from our simulations across reserve margins. The results are
shown in Figure 35 with stacked bars showing the costs customers would expect to pay on
average per kWh of consumption. The chart includes: (a) load-weighted-average energy costs,
(b) total capacity costs, calculated as the capacity price as paid to the entire ERCOT generation
and demand-response fleet at that reserve margin, and (c) an assumed transmission and
distribution cost rate, which would not vary with reserve margin.
The chart provides an indication of the impact on total customer rates across all customer classes.
We have not attempted to evaluate the differential cost impacts by customer class, nor for
customers with very different load profiles, nor for those that have demand-response capability.
In general, customers with higher load factors (flatter load profiles) will benefit less from energy
price reductions at higher reserve margins, but will also incur lower capacity costs per kWh of
load. Customers with some demand-response capability can further avoid incurring capacity
costs by either: (a) managing load away from peak conditions that might be used to assign
capacity payments, as in other RTOs with capacity markets and similar to how transmission rates
are allocated on a 4 coincident peak basis; or (b) explicitly counting those demand-response
capabilities on the supply side and earning capacity payments. The overall impact of increasing
the reserve margin will vary among customers, and could be calculated based on individual
assumed load profiles (although we have not attempted such a calculation here).
70 | brattle.com
Figure 35 Total Customer Costs on Average and in Highest 10% of Years 25¢
Infeasible Equilibria Increasing Capacity Payments
Supplier Net Revenues Associated with Progressively Higher Exceed Investment Costs Administratively‐Set Reserve Margins
Customer Costs (¢/kWh)
20¢
Energy‐Only Equilibrium 11.5% RM
Top 10% of Years
Unhedged
Top 10% of Years
80% Hedged
15¢
10¢
Energy
Capacity
5¢
Transmission and Distribution
0¢
7.9%
8.9%
9.8%
10.8% 11.8% 12.7% 13.7% 14.6% 15.6% 17.5%
Reserve Margin (ICAP %)
Sources and Notes: ERCOT Region T&D Costs of $42.8/MWh from EIA (2013), Table 55.1.
These results show that the probability-weighted average of annual customer costs are
minimized at exactly the energy-only equilibrium. Customer costs would be slightly higher at
either lower reserve margins or higher reserve margins. Lower reserve margins would lead to
higher average energy rates, although these would not persist for many years because suppliers
would earn returns in excess of CONE and therefore develop new resources in response. Higher
mandated reserve margins would reduce energy prices (and therefore customer costs), but would
require making capacity payments sufficient to fully compensate suppliers. This increase in
capacity costs is larger than the decrease in energy costs, resulting in a net increase in total
customer costs.
It is remarkable, however, to note how modest the customer cost increases are as the mandated
reserve margin increases.75 Higher mandated reserve margins with associated capacity payments
75
Costs increase faster on the left side of the curve at low reserve margins, but this observation can be
misleading since reserve margins below the 11.5% energy-only market equilibrium would not be
expected to occur on a 3-year forward basis. It is possible that 0-year forward reserve margins will
Continued on next page
71 | brattle.com
would cost customers only slightly more than the energy-only equilibrium. For example, the
14.1% reserve margin needed to maintain a 0.1 LOLE standard would cost customers only 0.1
¢/kWh more on average than the energy-only market at an 11.5% reserve margin in the longrun equilibrium. That represents an increase in average customer bills of only about 1%. This is
because the higher reserve margins reduce average annual energy prices, and the competitive
market sets capacity prices such that suppliers earn no more than CONE in total, no matter how
high the reserve margin requirement.
On an ERCOT-wide basis, raising the reserve margin from 11.5% to 14.1% would cost customers
approximately $400 million per year at equilibrium market conditions (from a $2.8 billion
reduction in energy costs offset by a $3.2 billion increase in capacity costs). However, these
estimates describe only long-term average prices at equilibrium. They do not describe this year
or the next few years. The actual near-term price impacts of implementing a capacity market
would be affected by at least two important dynamics. First, capacity prices could be temporarily
lower than estimated if some low-cost capacity is available. Other regions have experienced
capacity prices below Net CONE for many years due to the entry of demand response, generation
uprates, and other low-cost sources of capacity. Second, even without such resources, prices
would not be expected to reach equilibrium pricing until the reserve margin falls to the required
reserve margin. But herein lies an important cost difference from maintaining an energy-only
market. Mandating a reserve margin could cause the market to reach equilibrium as soon as
reserve margins fall from their current levels to 14.1% (or whatever level is mandated), whereas
the current energy-only market design may take several years to reach its 11.5% long-run
equilibrium level. Under either market design, long-term equilibrium prices would be several
billion dollars higher than currently-depressed wholesale prices, but a capacity requirement
would cause prices to reach equilibrium prices sooner, and at a slightly higher ultimate level
(e.g., $400 million higher on average to support a 0.1 LOLE).
Customers care about long-term average rates as well as year-to-year variability and uncertainty.
At a given planning reserve margin, wholesale spot prices for energy fluctuate because of
variations in load and generation availability (as well as gas price changes, which we have not
evaluated in this study). Capacity prices will also fluctuate with supply and demand conditions,
as explained in Section IV.B.2 above. The combined effect of energy and capacity price
uncertainty is reflected in the red dots in Figure 35, representing the average of retail prices for
the highest 10% of years, reflecting a once-per-decade scarcity year. As reserve margins
increase, this variability declines because the largest factor driving it is volatility in energy prices,
which make up a large portion of total customer costs. Volatility from uncertain capacity prices
increases with reserve margins, but this effect is less important since capacity costs are a
relatively smaller portion of total customer costs and because we assume that capacity prices are
capped at 2 times CONE.
Continued from previous page
occur, e.g., if load grows faster than expected, but this possibility is accounted for in the cost and
reliability distribution we show for the 3-year forward reserve margin of 11.5%.
72 | brattle.com
Most customers are not fully exposed to these spot price fluctuations, however, and are at least
partially hedged through fixed-price retail contracts and other arrangements. Hedging practices
vary by customer class and retail supplier. To illustrate the risk-mitigation effect of such
hedging, we assume that the average retail service will hedge 80% of weather-related price risk
(e.g., through seasonal forward contracts), but that the retailer will not hedge sufficiently far
forward to avoid any non-weather uncertainties such as load forecast error. We also assume that
customers would not be hedged against capacity price volatility, since we assume capacity prices
would be determined in three-year forward auctions and subsequently incorporated into retail
rates.
The results of this risk-mitigation analysis are shown in Figure 35 as pink dots, representing the
average annual customer rates during the highest 10% of all years. It shows that hedging
protects customers from the weather-driven extremes in energy spot prices that can occur at low
reserve margins. Hedging also reduces customer rate variability at higher reserve margins, but
not as much.
However, even when hedging practices are considered, customer rate variability will still decline
with increasing reserve margin. For example, the once-per-decade scarcity year would produce
prices of 15.1 ¢/kWh (about 50% more than average costs) under the energy-only market or
12.9 ¢/kWh (26% above average) under the capacity market at 0.1 LOLE. However, in either
case much of this once-per-decade volatility can be mitigated through hedging; a customer with
80% of energy purchases hedged on a seasonal basis and no capacity hedges would realize onceper-decade costs of 12.6 ¢/kWh (24% above average) under the energy-only market or
11.7 ¢/kWh (16% above average) under the capacity market.
2. Supplier Net Revenues and Volatility
Suppliers earn CONE under equilibrium market conditions under either a capacity market or an
energy-only market design, consistent with the incentives necessary to attract investments in a
deregulated electricity market. Reserve margins below the 11.5% energy-only equilibrium show
higher net revenues, but such conditions cannot persist because suppliers would earn margins
above CONE and invest in new resources until the reserve margin reached 11.5% and net
revenues dropped to CONE on average. Mandating reserve margins higher than 11.5% would
yield increasingly higher capacity prices to compensate for declining energy margins, as shown
in Figure 36. The individual components of the chart correspond to the customer cost chart
shown above, with the same assumptions about hedging against spot market uncertainty.
While supplier net revenues are equal across all reserve margins (above the minimum feasible
equilibrium at 11.5%), the proportion of those net revenues from the energy market declines
while those from the capacity market increase. For the gas CC reflected in the chart, capacity
payments increase from 0% of net revenues at the 11.5% energy-only equilibrium, up to 32% of
net revenues at the 14.1% reserve margin consistent with 1-in-10. The proportion of total
revenues from the capacity market would also vary by resource type, with baseload resources
earning most of their net revenues out of the energy market, while demand response and peakers
73 | brattle.com
would earn most revenues out of the capacity market. The increasing importance of capacity
payments to overall investment incentives at higher reserve margins highlights the importance of
carefully designing the capacity market so it produces efficient prices consistent with market
conditions.76
Figure 36 Supplier Net Revenues on Average and in Highest 10% of Years $800
Infeasible Equilibria Increasing Capacity Payments
Supplier Net Revenues Associated with Progressively Higher Exceed Investment Costs Administratively ‐Set Reserve Margins
Supplier Net Revenues ($/kW‐y)
$700
$600
Energy‐Only Equilibrium 11.5% RM
$500
$400
Top 10% of Years Unhedged
$300
Top 10% of Years 80% Hedged
$200
CC CONE Capacity Payments
Long‐Run Capacity
Price Needed to Sustain Reserve Margin
$100
Energy Margins
$0
7.9%
8.9%
9.8% 10.8% 11.8% 12.7% 13.7% 14.6% 15.6% 17.5%
Reserve Margin (ICAP %)
Similar to customers, suppliers experience more revenue volatility at lower reserve margins. We
illustrate these volatility impacts in Figure 36 by reporting net revenues in the once-per-decade
scarcity year if the supplier is totally exposed to spot energy and capacity prices (red dots) and,
alternatively, assuming hedging practices that eliminate approximately 80% of weather risks
(pink dots).77
76
We document best practices and pitfalls to avoid in capacity market design in Spees, et al. (2013) and
Pfeifenberger, et al. (2013).
77
See Section IV.C.1 for an explanation of how we implement the 80% seasonal energy hedging
assumption; see Section IV.B.2 for an explanation of how we estimate approximate capacity price
volatility.
74 | brattle.com
Because net revenues reflect the difference between prices and costs, supplier net revenues are
more volatile than customer rates. Similarly, the mitigating impacts of hedging and increasing
reserve margins are also greater for suppliers than they are for customers. For example, the onceper-decade scarcity year would produce supplier net revenues of $228/kW-year (1.9 times
CONE) in the energy-only market or $184/kW-year (1.5 times CONE) with a capacity market at
0.1 LOLE. If the supplier is 80% hedged, the once-per-decade year would produce net revenues
of $211/kW-year (1.7 times CONE) in the energy-only market or $172/kW-year (1.4 times
CONE) with a capacity market.
This reduced volatility at higher mandated reserve margins has implications for investor risk and,
as a likely result, suppliers’ cost of capital. Since electricity markets tend to be pro-cyclical (i.e.,
earnings increase when the overall economy expands and decrease in a poor economy), nonweather-related price risk is partly non-diversifiable and affects investors’ costs of capital and
therefore CONE. We have not quantified this effect, and so cannot evaluate its potential
magnitude. However, as an illustrative example calculation, if a higher reserve margin
requirement reduced investors’ cost of capital by, say, 10 basis points, CONE and all-in prices
would decrease slightly, by less than 1%. Hence, imposing higher reserve margin requirements
could at least marginally reduce the cost of capital and the market price for capacity, although we
have not attempted to quantify this effect and do not account for it in our analysis.
In addition, it is again important to note that Figure 36 represents supplier margins only under
equilibrium conditions. It does not account for possible transitional effects. Imposing a required
reserve margin could cause prices to climb from current levels toward equilibrium faster than
they would otherwise.
V. Policy Implications
The PUCT, ERCOT, and stakeholders have been addressing resource adequacy concerns since
2011, when extreme weather made the likely implications of a declining reserve margin trend
more tangible. Since then, the Commission held numerous workshops and sponsored several
studies aimed at addressing various aspects of the problem.78 The Commission has acted to
strengthen energy price signals by raising the price cap, implementing the ORDC, and other
reforms. Additional market design changes remain under discussion, including the possible
implementation of a mandatory reserve margin and an associated capacity market. These debates
have attracted considerable political interest. Meanwhile, suppliers have responded, at least
partially, to the increase in anticipated incentives by developing new plants.79
78
For access to a comprehensive set of associated studies and regulatory proceedings, see the Resource
Adequacy portion of ERCOT’s website as well as the PUCT proceedings under project 40,000. See
ERCOT (2014a) and PUCT (2014).
79
For example, Panda Power currently has three combined cycle stations under construction.
75 | brattle.com
Another significant development affecting the outlook for resource adequacy is ERCOT’s recent
load forecast update, based on a revised load forecasting methodology.80 Under the new
approach, ERCOT is now projecting substantially lower load growth compared to recent years.
If that forecast is accurate and no major retirements or project cancellations occur, ERCOT could
enjoy robust reserve margins for several years. That revised outlook may reduce the real or
perceived urgency of addressing resource adequacy concerns.
However, even if the immediacy of the concern is postponed for an intermediate number of
years, these issues highlight a complex set of difficult policy questions that must be answered for
the long term, including:
1. What reserve margin and supporting market design would be best for Texas?
2. What are the practical implications of maintaining the current energy-only
market design?
3. What would be the practical implications of introducing a resource adequacy
requirement and capacity market?
4. How would these conclusions change with market conditions, and what study
extensions or updates might be warranted?
We discuss here how our study results help to inform these policy questions as the Commission,
ERCOT, and stakeholders evaluate the most appropriate course for ERCOT’s market design for
resource adequacy.
1. What Reserve Margin and Supporting Market Design Would Be Best for Texas?
As we have explained previously, the most appropriate reserve margin and market design for
Texas depend on the policy objectives for resource adequacy that have yet to be fully articulated.
If economic efficiency is the only policy objective, then maintaining the energy-only market
design is likely the most appropriate course of action. As we have stated elsewhere and
illustrated through our Perfect Energy Price Case, a perfectly efficient energy-only market will
attract the economically optimal level of supply investments. In fact, our simulations indicate
that the current energy-only market design will sustain a reserve margin of approximately
11.5%, which is 1.3 percentage points above the risk-neutral, economically optimum level of
approximately 10.2%, because the ORDC curve will sometimes produce prices in excess of
marginal system cost.
However, the Commission must weigh multiple, sometimes conflicting, policy objectives that
might potentially be best supported by a mandatory reserve margin requirement, as implemented
in a well-designed capacity market. Specifically, implementing a capacity market would reduce
the risks associated with potential low-reliability and high-cost events, providing net benefits
80
See ERCOT (2014b).
76 | brattle.com
overall from a risk-averse (rather than risk-neutral) perspective. Only by carefully examining
the reliability, economic, and other policy implications of increasing reserve margins can the
Commission determine the market design that will best serve the state of Texas.
If policymakers and stakeholders place a greater weight on the potential high-cost and lowreliability outcomes that can result from extreme weather, unexpectedly high load growth,
unusual generation outages, or modelling uncertainties, then a higher mandated reserve margin
could be justified. For example, at the 11.5% energy-only equilibrium in the Base Case, the
LOLE would be about 0.3 events per year, which would result in an average of about one load
shedding event due to inadequate installed generation capacity every three to four years. Each
such event would shed about 1,600 MW of load (approximately 2% of peak load) for about 2.6
hours. At an assumed VOLL of $9,000/MWh, the implied cost of these load-shed events is
approximately $40 million. This economic value is already incorporated as one of the
components of our analysis and reflected in the 10.2% risk-neutral, economically optimal reserve
margin, but the cost of such “blackout” events may appear to be much higher from a public,
political, and regulatory policy perspective.
Strong aversion to load shed events is difficult to assess from an economic perspective. However,
potentially strong risk aversion preferences are suggested by the level of press attention and
public reaction that follows rolling blackout events. A higher mandated reserve margin would
reduce the risk of such events. It would also pre-empt the possible regulatory and legislative
interventions in ERCOT’s power market that might follow such an event, the prospect of which
increases perceived regulatory risk by suppliers and reduces their incentive to invest in an
energy-only market.
A high level of risk aversion to load shed or extreme high price events may justify the additional
cost of a maintaining a high reserve margin. One of the most interesting findings from this study
is the slow rate at which total system costs and average customer costs would likely increase with
higher reserve margins. Because the increased expense associated with capacity payments is
partially offset by reduced energy market prices and reduced costs of reliability events, the net
cost increase associated with mandating reserve margins above the energy-only equilibrium is
just slightly higher on a long-term average basis. For example, our simulation results suggest that
the net increase in the average system-wide costs of increasing the reserve margin from 11.5% to
14.1% would be in the order of $100 million per year. If risk aversion to load shed events is very
high, these costs may be justified to support a higher reserve margin.
Some market participants have suggested the possibility of implementing a capacity market with
a reserve margin mandate set at the risk-neutral, economic optimum; another option would be to
increase that reserve margin slightly to reflect some level of risk aversion. At the risk-neutral
optimum, this would mean imposing a 10.2% minimum reserve margin requirement, which is
below the likely energy-only market equilibrium of 11.5% (a relationship that may not have
been anticipated by the market participants suggesting this possibility). While this approach
would not increase the average planning reserve margin, it still has theoretical merit.
Implementing a 10.2% or 11.5% reserve margin would not increase system costs compared to an
77 | brattle.com
energy-only market at the same reserve margin, but would still provide some of the benefits of a
capacity market, including: (a) reducing the risk of and high costs of low-reliability outcomes
under which reserve margins fall below the mandated reserve margin; (b) providing
transparency that would help rationalize supply and demand outlooks on a three-year forward
basis, thereby reducing the likelihood of boom-bust cycles; (c) providing a more stable price
signal for maintaining the chosen reserve margin; and (d) creating a competitive centralized
auction within which all types of supply, including demand response, can compete and possibly
improve the overall efficiency of investment decisions in the fleet.
However, implementing mandated reserve margins and an associated capacity market will add
considerable complexity to ERCOT’s current energy market design. This complexity may not be
justified if the current design already sustains an equilibrium reserve margin of 11.5%, which is
above our 10.2% estimate of the risk-neutral, economically optimal reserve margin. Costs also
include having to design, implement, and administer a complex new market with many
administrative determinations that will have significant economic consequences and are likely to
be litigated.
2. What Are the Practical Implications of Maintaining the Current Energy-Only Market
Design?
Maintaining the current energy-only market would avoid the complexity of introducing major
new market design elements. The Commission and ERCOT would continue with their various
ongoing initiatives to enhance ERCOT’s existing energy and ancillary service markets to support
reliable operations with maximal economic efficiency. In particular, we recommend continuing
efforts to enable the economic and efficient participation of demand-response resources. The
Commission might consider further assessing whether there are ways to further improve the
ability of demand response to participate efficiently in the energy and ancillary services markets.
However, it may be beneficial to codify the overarching market design principles to discourage
future interventions in the market whenever prices spike or rare reliability events occur,
recognizing that such outcomes are an inherent component of the chosen energy-only market
design. Without such regulatory assurances, potential investors may discount the expected
revenues that could be realized under these shortage conditions. Articulating more clearly the
principles and expectations supporting the current design may reduce this regulatory uncertainty
and help attract the needed investments in the energy-only market.
3. What Would Be the Practical Implications of Introducing a Resource Adequacy
Requirement and Capacity Market?
Establishing a resource adequacy requirement and a capacity market would be a major market
design effort. It would require extensive stakeholder discussions about market design details,
including the determination of a number of administrative parameters. Design elements that
would have to be addressed include: (1) the resource adequacy requirement itself, which could
be based on various economic or reliability standards as discussed in this report; (2) the
78 | brattle.com
implementation of that requirement in a capacity market, which could involve administratively
defining a sloped demand curve; (3) the forward period and rules for forward and incremental
auctions; (4) whether and how to represent transmission constraints; (5) participation and
verification rules for all types of resources; (6) definition of penalties and performance incentives;
(7) market monitoring protocols to preventing the exercise of market power by suppliers or
manipulation by buyers; and (8) settlement processes and rules for both suppliers and loadserving entities.
Regions that have implemented capacity markets have spent years developing these design
elements and have been adjusting them periodically, often in a litigation setting. Many of these
design elements also become politically charged, particularly if policymakers believe customer
costs have increased inefficiently. ERCOT and the Commission would face similar challenges if
they adopted a capacity market, although it could learn from the other regions’ experiences.
They could avoid some of the other regions’ complications, such as multi-state jurisdictions and
interactions with FERC oversight.
One challenge is the transition period when first implementing a capacity market. Specifically,
mandating a higher reserve margin could cause a significant increase in total customer costs over
the next few years relative to the energy-only course, with these near-term cost increases
exceeding the long-run equilibrium cost differential that we have estimated in this study. Shortterm cost increases would be driven by the fact that the ERCOT system as it stands may have
only a small excess relative to the capacity market equilibrium reserve margin (meaning that
equilibrium price levels may be reached very soon). Meanwhile the current reserve margin
exceeds the energy-only equilibrium reserve margin by several percentage points, and so energyonly market prices may not reach higher long-run sustainable levels for several more years.
Other transitional challenges include phasing in procurement of capacity to achieve a three- or
four-year forward procurement, addressing legacy contracts and resources, and specifying the
resource adequacy values of different resource types, including intermittent resources,
uncommitted supply, and interties with neighboring markets.
Another concern about implementing a resource adequacy requirement and an associated
capacity market is that it would introduce significant complexity without directly addressing all
aspects of reliability. The resource adequacy discussion is focused on ensuring that installed
generation capacity is sufficient to meet load at summer peak. But this is only one aspect of
system reliability and does not address the much more frequent distribution system-related
outages. Even from a system reliability perspective, gas supply disruptions or widespread
freezing and drought conditions can disable supply and lead to customer outages despite
adequate levels of installed reserve margins. Outages may also result from inadequate ramping
capability for meeting rapid changes in load and wind generation, and other operational
challenges. The ORDC design already addresses some of these challenges by providing incentive
to invest slightly beyond the risk-neutral, economically optimal level of installed capacity and
operational performance.
79 | brattle.com
4. How Would These Conclusions Change with Market Conditions, and What Study
Extensions or Updates Might Be Warranted?
The quantitative analysis presented in this study is based on many data inputs, simulations to
approximate projected system characteristics, assumed market conditions, and market rules
expected for 2016. If the PUCT were to mandate a particular reserve margin based on the
analysis presented in this study, it would be necessary to update the analysis as system
conditions, market conditions, and market rules change in the future. For example, a new load
forecast with a wider (or narrower) distribution of possible load-growth outcomes could increase
(or decrease) the estimated energy-only equilibrium and the optimal reserve margins. A higher
(or lower) gas price forecast could similarly increase (or decrease) the reserve margin results.
Any change in accounting for the resource adequacy value of transmission ties with neighboring
markets, wind power plant, or demand response would also change the level of specified reserve
margins.
We recognize that concurrent with the completion of our study, ERCOT has released an updated
load forecast that we have not had time to incorporate and consider in our analyses. Because the
updated forecast is much lower than prior forecasts, this may reduce the real or perceived
urgency of addressing the resource adequacy question. In terms of the impact on our study
results, we do not expect that the new forecast would change our estimates of optimal or
equilibrium reserve margins substantially, since those are expressed as a percentage of peak load.
However, our results could change if the new forecast methodology produces very different
distributions of weather and non-weather forecast errors, or if it accounts for demand response
very differently.
Regarding the urgency of the issue, that is a judgment for the Commission, ERCOT, and
stakeholders. The Commission could decide not to act now or, if they see a need to change the
market design in the long term to meet policy objectives, they could take advantage of the
current slack to implement changes with less risk of transitional rate shocks. In any case,
providing stakeholders a clear understanding of whether, how, and when the market design will
change would reduce regulatory uncertainty and benefit market participants as they make
business decisions over the coming years.
80 | brattle.com
List of Acronyms
1-in-10
1-Day-In-Ten-Years, which can refer to either 1 load shed event in 10 years or 24
hours of load shedding in 10 years
4 CP
Four Coincident Peak
A/S
Ancillary Service
ATC
Available Transfer Capability
ATWACC
After-Tax Weighted-Average Cost of Capital
Btu
British Thermal Unit
CC
Combined Cycle
CDR
Capacity, Demand, and Reserves
CONE
Cost of New Entry
CT
Combustion Turbine
DC
Direct Current
DR
Demand Response
EE
Energy Efficiency
EEA
Energy Emergency Alert
ELCC
Effective Load Carrying Capability
ERCOT
Electric Reliability Council of Texas
ERS
Emergency Response Service
EUE
Expected Unserved Energy
FERC
Federal Energy Regulatory Commission
GADS
Generation Availability Data System
HCAP
High System-Wide Offer Cap
HHV
Higher Heating Value
81 | brattle.com
HVDC
High Voltage Direct Current
ICAP
Installed Capacity
ISO
Independent System Operator
kW
Kilowatt
kWh
Kilowatt hour
LCAP
Low System-Wide Offer Cap
LFE
Load Forecast Error
LOLE
Loss of Load Event
LOLH
Loss of Load Hours
LOLP
Loss of Load Probability
LR
Load Resource
MISO
Midcontinent Independent System Operator
MMBtu
One Million British Thermal Units
MW
Megawatt
MWh
Megawatt Hour
NERC
North American Electric Reliability Corporation
NSRS
Non-Spinning Reserve Service
NYISO
New York Independent System Operator
ORDC
Operating Reserve Demand Curve
PBPC
Power Balance Penalty Curve
PNM
Peaker Net Margin
PRC
Physical Responsive Capability
PRD
Price-Responsive Demand
PUCT
Public Utility Commission of Texas
82 | brattle.com
PUN
Private Use Network
RM
Reserve Margin
RRS
Responsive Reserve Service
RT
Real-Time
SCED
Security Constrained Economic Dispatch
SERVM
Strategic Energy Risk Valuation Model
SPP
Southwest Power Pool
ST
Steam Turbine
SWOC
System-Wide Offer Cap
T&D
Transmission and Distribution
TDSP
Transmission and Distribution Service Provider
UCAP
Unforced Capacity
VOLL
Value of Lost Load
VOM
Variable Operations and Maintenance
83 | brattle.com
Bibliography
Astrape Consulting (2014), SERVM Software: Strategic Energy & Risk Valuation Model
(SERVM). (Astrape 2014) Available at http://www.astrape.com/servm/
Australia Energy Market Commission (2007). Comprehensive Reliability Review: Final Report.
December 2007. Retrieved from http://www.aemc.gov.au/marketreviews/completed/comprehensive-reliability-review.html
Australia Energy Market Commission (2010). Reliability Standard and Reliability Settings
Review. April 30, 2010. Retrieved from
http://www.aemc.gov.au/Media/docs/Final%20Report%20%20Reliability%20Standard%20and%20Settings%20Review-c3994458-70b3-4d66-badddc7e8b809f68-1.PDF
Bloomberg Terminal. (2014) Data downloaded over October 2013 to January 2014.
Carden, Kevin, Nick Wintermantel, and Johannes Pfeifenberger (2011). The Economics of
Resource Adequacy Planning: Why Reserve Margins Are Not Just About Keeping the Lights
On. April 2011. Available: http://www.cbo.gov/sites/default/files/cbofiles/attachments/01-312012_Outlook.pdf
ECCO International, Inc. (2013a), 2013 ERCOT Loss of Load Study Assumptions and
Methodology, March 7, 2013. (ECCO, 2013) Available at
http://www.ercot.com/content/gridinfo/resource/2014/reservemargin/ERCOT_Loss_of_Load
_Study_2013.pdf
ECCO International, Inc. (2013b), 2013 ERCOT Loss of Load STUDY Results, March 8, 2013.
(ECCO, 2013b) Available at
http://www.ercot.com/content/gridinfo/resource/2014/reservemargin/ERCOT_Loss_of_Load
_Study_2013PartII.pdf
Electric Reliability Council of Texas (ERCOT) (2013a). Report on the Capacity, Demand, and
Reserves in the ERCOT Region, May 2013. (May 2013 CDR) Available at
http://www.ercot.com/content/gridinfo/resource/2014/adequacy/cdr/Capacity_DemandandR
eserveReport-May2013.pdf
Electric Reliability Council of Texas (ERCOT) (2013b). ERCOT Business Practice: Setting the
Shadow Price Caps and Power Balance Penalties in Security Constrained Economic Dispatch,
Public, Version_8.0, effective July 17, 2013. (ERCOT 2013b) Available at
http://www.ercot.com/content/mktinfo/rtm/kd/Methodology%20for%20Setting%20Maximu
m%20Shadow%20Prices%20for%20Network%20an.zip
84 | brattle.com
Electric Reliability Council of Texas (ERCOT) and William Hogan (2013c). Back Cast of Interim
Solution B+ to Improve Real-Time Scarcity Pricing, White Paper, March 21, 2013. Available
at
http://www.ercot.com/content/gridinfo/resource/2014/mktanalysis/White_Paper_Back_Cast_
of_Interim_Solution_B+_to_Improve_Re.pdf
Electric Reliability Council of Texas (ERCOT) (2013d). ERCOT Nodal Protocols, Public,
October 7, 2013. (ERCOT, 2013d) Available at
http://www.ercot.com/content/mktrules/nprotocols/2013/10a/October_7,_2013_Nodal_Proto
cols.pdf
Electric Reliability Council of Texas (ERCOT) (2013e). Demand Response: Emergency
Response Service. (ERCOT, 2013e) Available at
http://www.ercot.com/services/programs/load/eils/
Electric Reliability Council of Texas (ERCOT) (2013f). Emergency Response Service: Technical
Requirements and Scope of Work – Standard Contract Term: October 1, 2013 through
January 31, 2014, August 12, 2013 (ERCOT, 2103f) Available at
http://www.ercot.com/services/programs/load/eils/
Electric Reliability Council of Texas (ERCOT) (2013g). Pilot Projects: 30-Minute Emergency
Response Service: 30-Minute ERS Governing Document, July 26, 2013. (ERCOT, 2013g)
Available at http://www.ercot.com/mktrules/pilots/ers/index
Electric Reliability Council of Texas (ERCOT) (2013h). Market Guides: Nodal Operating
Guides, Online at http://www.ercot.com/mktrules/guides/noperating/
Electric Reliability Council of Texas (ERCOT) (2014a), Grid Information: Resource Adequacy.
(ERCOT 2014a) Available at http://www.ercot.com/gridinfo/resource/
Electric Reliability Council of Texas (ERCOT) (2014b), Review of Preliminary Load Forecast
Error Statistics Summary, January 27, 2014, Workshop–Review CDR Load Forecast
Methodology. Available at http://www.ercot.com/calendar/2014/01/20140127CDRWorkshop
Federal Energy Regulatory Commission (FERC) (2013a). Form No. 714—Annual Electric
Balancing Authority Area and Planning Area Report: 2006–2012 Form 714 Database, August
5, 2013. (FERC 714, 2013) Available at https://www.ferc.gov/docs-filing/forms/form714/data.asp
Federal Energy Regulatory Commission (FERC) (2013b).
Munasinghe, M. and A. Sanghvi (1988). “Reliability of Supply, Outage Costs and Value of
Service: An Overview,” The Energy Journal. Volume 9, Special Issue on Electricity
Reliability.
85 | brattle.com
National Oceanic and Atmospheric Administration (NOAA) (2013). National Climatic Data
Center, Climate Data Online. (NOAA, 2013) Available at http://www.ncdc.noaa.gov/cdoweb/
Newell, Samuel, et al. (2012), ERCOT Investment Incentives and Resource Adequacy, prepared
for ERCOT, June 1, 2012. Available at
http://www.brattle.com/system/publications/pdfs/000/004/820/original/ERCOT_Investment_
Incentives_and_Resource_Adequacy_Newell_Spees_Pfeifenberger_Mudge_ERCOT_June_2_
2012.pdf?1378772132
Newell, Sam and Mike DeLucia (2013) “Additional ORDC B+ Economic Equilibrium Planning
Reserve Margin Estimates,” prepared for The Public Utility Commission of Texas, July 2013,
Attachment B to Responses to Commission’s Request for Analyses of Additional ORDC B+
Scenarios, in the Mater of Commission Proceeding to Ensure Resource Adequacy, Project No.
40000, before the Public Utility Commission of Texas. Available at
http://interchange.puc.state.tx.us/WebApp/Interchange/application/dbapps/filings/pgSearch_
Results.asp?TXT_CNTR_NO=40000&TXT_ITEM_NO=437
Newell, Samuel A., J. Michael Hagerty, and Quincy X. Liao. (2013b). 2013 Offer Review Trigger
Prices Study. Prepared for the ISO New England. October 2013. Filed before the Federal
Energy Regulatory Commission in Docket ER14-616.
Newell, Samuel A., Kathleen Spees, Michael DeLucia, and Ben Housman. (2014) “Draft
Recommendations for a Capacity Demand Curve in ISO-NE.” January 14, 2014. Available:
http://www.isone.com/committees/comm_wkgrps/mrkts_comm/mrkts/mtrls/2014/jan14152014/a04_the_bra
ttle_group_capacity_demand_curve_presentation_01_14_14.pptx
Nordel (2009). Guidelines for the Implementation of Transitional Peak Load Arrangements:
Proposal of Nordel. March 2009. Retrieved from
http://www.nordpoolspot.com/Global/Download%20Center/TSO/entsoe_nordic-guidelinesfor-implementation-of-transitional-peak-load-arrangements.pdf
North American Electric Reliability Corporation (NERC) (2010). G&T Reliability Planning
Models Task Force (GTRPMTF): Methodology and Metrics, Revised October 15, 2010.
Retrieved from
http://www.wecc.biz/committees/StandingCommittees/PCC/LRS/01062011/Lists/Minutes/1/
GTRPMTF%20Final%20PCapproved%20Methodology%20and%20Metrics%20redline%2029Oct10%20mods.pdf
North American Electric Reliability Corporation (NERC) (2012). 2012 Long-Term Reliability
Assessment, November 2012. (NERC 2012)
Available at NERC 2012 to bib:
http://www.nerc.com/files/2012_ltra_final.pdf
86 | brattle.com
OATI webOasis Production Node (2013). Southwest Power Pool OASIS. (OATI 2013).
Available at http://www.oasis.oati.com/SWPP/index.html
Pfeifenberger, Johannes P., Kathleen Spees, and Adam Schumacher. (2009) A Comparison of
PJM’s RPM with Alternative Energy and Capacity Market Designs. September 2009.
Available:
http://www.brattle.com/system/publications/pdfs/000/004/859/original/A_Comparison_of_PJ
M%E2%80%99s_RPM_with_Alt_Energy_and_Capacity_Mkt_Designs_Pfeifenberger_et_al_S
ep_2009.pdf?1379014789
Pfeifenberger, Johannes P., Kathleen Spees, and Samuel A. Newell (2012). Resource Adequacy
in California: Options for Improving Efficiency and Effectiveness, October 2012. Available at
http://www.brattle.com/system/publications/pdfs/000/004/827/original/Resource_Adequacy_i
n_California_Calpine_Pfeifenberger_Spees_Newell_Oct_2012.pdf?1378772133
Pfeifenberger, Johannes and Kathleen Spees (2013). “Characteristics of Successful Capacity
Markets,” presented at APEx Conference 2013, New York, NY, October 31, 2013. Available
at
http://www.brattle.com/system/publications/pdfs/000/004/951/original/Characteristics_of_Su
ccessful_Capacity_Markets_Pfeifenberger_Spees_Oct_2013.pdf?1383246105
Poland, William B. (1988). “The Importance of Including Uncertainties in Economic Generation
Reliability Planning,” The Energy Journal. Volume 9, Special Issue on Electricity Reliability.
Public Utility Commission of Texas (PUCT) (2012). Chapter 25. Substantive Rules Applicable to
Electric Service Providers. Subchapter S. Wholesale Markets, Effective Date November 15,
2012. (PUCT 2014b) Available at
http://www.puc.texas.gov/agency/rulesnlaws/subrules/electric/25.505/25.505.pdf
Public Utility Commission of Texas (PUCT) (2014), Commission Proceeding to Ensure Resource
Adequacy in Texas, Project 40,000. (PUCT 2014) Available at
http://interchange.puc.state.tx.us/WebApp/Interchange/application/dbapps/filings/pgControl.
asp?TXT_UTILITY_TYPE=A&TXT_CNTRL_NO=40000&TXT_ITEM_MATCH=1&TXT_ITE
M_NO=&TXT_N_UTILITY=&TXT_N_FILE_PARTY=&TXT_DOC_TYPE=ALL&TXT_D_FR
OM=&TXT_D_TO=&TXT_NEW=true
SNL Energy. (2014) Data downloaded over October 2013 to January 2014.
Southwest Power Pool (2010). 2010 Loss of Load Expectation Report. October 8, 2010. Retrieved
from http://www.spp.org/publications/LOLE%20Report_5%20Draft_cc.pdf
Spees, Kathleen, Samuel A. Newell, Robert Carlton, Bin Zhou, and Johannes P. Pfeifenberger.
(2011) Cost of New Entry Estimates for Combustion Turbine and Combined-Cycle Plants in
PJM. August 24, 2011.
87 | brattle.com
Spees, Kathleen, Samuel A. Newell, and Johannes P. Pfeifenberger. (2013). “Capacity Markets:
Lessons Learned from the First Decade,” Economics of Energy & Environmental Policy. Vol.
2, No. 2.
U.S. Energy Information Administration (EIA) (2013). Annual Energy Outlook 2013. Table
55.1. “Electric Power Projections for EMM Region, Texas Regional Entity, Reference Case.
(EIA 2013) Retrieved from
http://www.eia.gov/oiaf/aeo/tablebrowser/#release=AEO2013&subject=0AEO2013&table=62-AEO2013&region=3-1&cases=ref2013-d102312a
U.S. Energy Information Administration (EIA) (2014). Form EIA-923 detailed date, 2014).
Available at EIA 923, http://www.eia.gov/electricity/data/eia923/
Ventyx Energy Velocity Suite. (2014) Data downloaded over October 2013 to January 2014.
88 | brattle.com